source: doc/papers/concurrency/Paper.tex @ 62dbb00

arm-ehjacob/cs343-translationjenkins-sandboxnew-astnew-ast-unique-expr
Last change on this file since 62dbb00 was 62dbb00, checked in by Peter A. Buhr <pabuhr@…>, 2 years ago

add referees for SP&E submission, start new mail file

  • Property mode set to 100644
File size: 156.2 KB
Line 
1\documentclass[AMA,STIX1COL]{WileyNJD-v2}
2
3\articletype{RESEARCH ARTICLE}%
4
5% Referees
6% Doug Lea, dl@cs.oswego.edu, SUNY Oswego
7% Herb Sutter, hsutter@microsoft.com, Microsoft Corp
8% Gor Nishanov, gorn@microsoft.com, Microsoft Corp
9% James Noble, kjx@ecs.vuw.ac.nz, Victoria University of Wellington, School of Engineering and Computer Science
10
11\received{XXXXX}
12\revised{XXXXX}
13\accepted{XXXXX}
14
15\raggedbottom
16
17%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
18
19% Latex packages used in the document.
20
21\usepackage{epic,eepic}
22\usepackage{xspace}
23\usepackage{enumitem}
24\usepackage{comment}
25\usepackage{upquote}                                            % switch curled `'" to straight
26\usepackage{listings}                                           % format program code
27\usepackage[labelformat=simple,aboveskip=0pt,farskip=0pt]{subfig}
28\renewcommand{\thesubfigure}{(\Alph{subfigure})}
29\captionsetup{justification=raggedright,singlelinecheck=false}
30\usepackage{dcolumn}                                            % align decimal points in tables
31\usepackage{capt-of}
32\setlength{\multicolsep}{6.0pt plus 2.0pt minus 1.5pt}
33
34\hypersetup{breaklinks=true}
35\definecolor{OliveGreen}{cmyk}{0.64 0 0.95 0.40}
36\definecolor{Mahogany}{cmyk}{0 0.85 0.87 0.35}
37\definecolor{Plum}{cmyk}{0.50 1 0 0}
38
39\usepackage[pagewise]{lineno}
40\renewcommand{\linenumberfont}{\scriptsize\sffamily}
41
42\renewcommand{\topfraction}{0.8}                        % float must be greater than X of the page before it is forced onto its own page
43\renewcommand{\bottomfraction}{0.8}                     % float must be greater than X of the page before it is forced onto its own page
44\renewcommand{\floatpagefraction}{0.8}          % float must be greater than X of the page before it is forced onto its own page
45\renewcommand{\textfraction}{0.0}                       % the entire page maybe devoted to floats with no text on the page at all
46
47\lefthyphenmin=3                                                        % hyphen only after 4 characters
48\righthyphenmin=3
49
50%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
51
52% Names used in the document.
53
54\newcommand{\CFAIcon}{\textsf{C}\raisebox{\depth}{\rotatebox{180}{\textsf{A}}}\xspace} % Cforall symbolic name
55\newcommand{\CFA}{\protect\CFAIcon}             % safe for section/caption
56\newcommand{\CFL}{\textrm{Cforall}\xspace}      % Cforall symbolic name
57\newcommand{\Celeven}{\textrm{C11}\xspace}      % C11 symbolic name
58\newcommand{\CC}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}\xspace} % C++ symbolic name
59\newcommand{\CCeleven}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}11\xspace} % C++11 symbolic name
60\newcommand{\CCfourteen}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}14\xspace} % C++14 symbolic name
61\newcommand{\CCseventeen}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}17\xspace} % C++17 symbolic name
62\newcommand{\CCtwenty}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}20\xspace} % C++20 symbolic name
63\newcommand{\Csharp}{C\raisebox{-0.7ex}{\Large$^\sharp$}\xspace} % C# symbolic name
64
65%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
66
67\newcommand{\Textbf}[2][red]{{\color{#1}{\textbf{#2}}}}
68\newcommand{\Emph}[2][red]{{\color{#1}\textbf{\emph{#2}}}}
69\newcommand{\uC}{$\mu$\CC}
70\newcommand{\TODO}[1]{{\Textbf{#1}}}
71
72%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
73
74% Default underscore is too low and wide. Cannot use lstlisting "literate" as replacing underscore
75% removes it as a variable-name character so keywords in variables are highlighted. MUST APPEAR
76% AFTER HYPERREF.
77%\DeclareTextCommandDefault{\textunderscore}{\leavevmode\makebox[1.2ex][c]{\rule{1ex}{0.1ex}}}
78\renewcommand{\textunderscore}{\leavevmode\makebox[1.2ex][c]{\rule{1ex}{0.075ex}}}
79
80\renewcommand*{\thefootnote}{\Alph{footnote}} % hack because fnsymbol does not work
81%\renewcommand*{\thefootnote}{\fnsymbol{footnote}}
82
83\makeatletter
84% parindent is relative, i.e., toggled on/off in environments like itemize, so store the value for
85% use rather than use \parident directly.
86\newlength{\parindentlnth}
87\setlength{\parindentlnth}{\parindent}
88
89\newcommand{\LstBasicStyle}[1]{{\lst@basicstyle{\lst@basicstyle{#1}}}}
90\newcommand{\LstKeywordStyle}[1]{{\lst@basicstyle{\lst@keywordstyle{#1}}}}
91\newcommand{\LstCommentStyle}[1]{{\lst@basicstyle{\lst@commentstyle{#1}}}}
92
93\newlength{\gcolumnposn}                                        % temporary hack because lstlisting does not handle tabs correctly
94\newlength{\columnposn}
95\setlength{\gcolumnposn}{3.5in}
96\setlength{\columnposn}{\gcolumnposn}
97
98\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}}}}
99\newcommand{\CRT}{\global\columnposn=\gcolumnposn}
100
101% Denote newterms in particular font and index them without particular font and in lowercase, e.g., \newterm{abc}.
102% The option parameter provides an index term different from the new term, e.g., \newterm[\texttt{abc}]{abc}
103% The star version does not lowercase the index information, e.g., \newterm*{IBM}.
104\newcommand{\newtermFontInline}{\emph}
105\newcommand{\newterm}{\@ifstar\@snewterm\@newterm}
106\newcommand{\@newterm}[2][\@empty]{\lowercase{\def\temp{#2}}{\newtermFontInline{#2}}\ifx#1\@empty\index{\temp}\else\index{#1@{\protect#2}}\fi}
107\newcommand{\@snewterm}[2][\@empty]{{\newtermFontInline{#2}}\ifx#1\@empty\index{#2}\else\index{#1@{\protect#2}}\fi}
108
109% Latin abbreviation
110\newcommand{\abbrevFont}{\textit}                       % set empty for no italics
111\@ifundefined{eg}{
112\newcommand{\EG}{\abbrevFont{e}\abbrevFont{g}}
113\newcommand*{\eg}{%
114        \@ifnextchar{,}{\EG}%
115                {\@ifnextchar{:}{\EG}%
116                        {\EG,\xspace}}%
117}}{}%
118\@ifundefined{ie}{
119\newcommand{\IE}{\abbrevFont{i}\abbrevFont{e}}
120\newcommand*{\ie}{%
121        \@ifnextchar{,}{\IE}%
122                {\@ifnextchar{:}{\IE}%
123                        {\IE,\xspace}}%
124}}{}%
125\@ifundefined{etc}{
126\newcommand{\ETC}{\abbrevFont{etc}}
127\newcommand*{\etc}{%
128        \@ifnextchar{.}{\ETC}%
129        {\ETC.\xspace}%
130}}{}%
131\@ifundefined{etal}{
132\newcommand{\ETAL}{\abbrevFont{et}~\abbrevFont{al}}
133\newcommand*{\etal}{%
134        \@ifnextchar{.}{\protect\ETAL}%
135                {\protect\ETAL.\xspace}%
136}}{}%
137\@ifundefined{viz}{
138\newcommand{\VIZ}{\abbrevFont{viz}}
139\newcommand*{\viz}{%
140        \@ifnextchar{.}{\VIZ}%
141                {\VIZ.\xspace}%
142}}{}%
143\makeatother
144
145\newenvironment{cquote}
146               {\list{}{\lstset{resetmargins=true,aboveskip=0pt,belowskip=0pt}\topsep=3pt\parsep=0pt\leftmargin=\parindentlnth\rightmargin\leftmargin}%
147                \item\relax}
148               {\endlist}
149
150%\newenvironment{cquote}{%
151%\list{}{\lstset{resetmargins=true,aboveskip=0pt,belowskip=0pt}\topsep=3pt\parsep=0pt\leftmargin=\parindentlnth\rightmargin\leftmargin}%
152%\item\relax%
153%}{%
154%\endlist%
155%}% cquote
156
157% CFA programming language, based on ANSI C (with some gcc additions)
158\lstdefinelanguage{CFA}[ANSI]{C}{
159        morekeywords={
160                _Alignas, _Alignof, __alignof, __alignof__, asm, __asm, __asm__, __attribute, __attribute__,
161                auto, _Bool, catch, catchResume, choose, _Complex, __complex, __complex__, __const, __const__,
162                coroutine, disable, dtype, enable, exception, __extension__, fallthrough, fallthru, finally,
163                __float80, float80, __float128, float128, forall, ftype, generator, _Generic, _Imaginary, __imag, __imag__,
164                inline, __inline, __inline__, __int128, int128, __label__, monitor, mutex, _Noreturn, one_t, or,
165                otype, restrict, __restrict, __restrict__, __signed, __signed__, _Static_assert, thread,
166                _Thread_local, throw, throwResume, timeout, trait, try, ttype, typeof, __typeof, __typeof__,
167                virtual, __volatile, __volatile__, waitfor, when, with, zero_t},
168        moredirectives={defined,include_next},
169        % replace/adjust listing characters that look bad in sanserif
170        literate={-}{\makebox[1ex][c]{\raisebox{0.4ex}{\rule{0.8ex}{0.1ex}}}}1 {^}{\raisebox{0.6ex}{$\scriptstyle\land\,$}}1
171                {~}{\raisebox{0.3ex}{$\scriptstyle\sim\,$}}1 % {`}{\ttfamily\upshape\hspace*{-0.1ex}`}1
172                {<}{\textrm{\textless}}1 {>}{\textrm{\textgreater}}1
173                {<-}{$\leftarrow$}2 {=>}{$\Rightarrow$}2 {->}{\makebox[1ex][c]{\raisebox{0.5ex}{\rule{0.8ex}{0.075ex}}}\kern-0.2ex{\textrm{\textgreater}}}2,
174}
175
176\lstset{
177language=CFA,
178columns=fullflexible,
179basicstyle=\linespread{0.9}\sf,                                                 % reduce line spacing and use sanserif font
180stringstyle=\tt,                                                                                % use typewriter font
181tabsize=5,                                                                                              % N space tabbing
182xleftmargin=\parindentlnth,                                                             % indent code to paragraph indentation
183%mathescape=true,                                                                               % LaTeX math escape in CFA code $...$
184escapechar=\$,                                                                                  % LaTeX escape in CFA code
185keepspaces=true,                                                                                %
186showstringspaces=false,                                                                 % do not show spaces with cup
187showlines=true,                                                                                 % show blank lines at end of code
188aboveskip=4pt,                                                                                  % spacing above/below code block
189belowskip=3pt,
190moredelim=**[is][\color{red}]{`}{`},
191}% lstset
192
193% uC++ programming language, based on ANSI C++
194\lstdefinelanguage{uC++}[ANSI]{C++}{
195        morekeywords={
196                _Accept, _AcceptReturn, _AcceptWait, _Actor, _At, _CatchResume, _Cormonitor, _Coroutine, _Disable,
197                _Else, _Enable, _Event, _Finally, _Monitor, _Mutex, _Nomutex, _PeriodicTask, _RealTimeTask,
198                _Resume, _Select, _SporadicTask, _Task, _Timeout, _When, _With, _Throw},
199}
200\lstdefinelanguage{Golang}{
201        morekeywords=[1]{package,import,func,type,struct,return,defer,panic,recover,select,var,const,iota,},
202        morekeywords=[2]{string,uint,uint8,uint16,uint32,uint64,int,int8,int16,int32,int64,
203                bool,float32,float64,complex64,complex128,byte,rune,uintptr, error,interface},
204        morekeywords=[3]{map,slice,make,new,nil,len,cap,copy,close,true,false,delete,append,real,imag,complex,chan,},
205        morekeywords=[4]{for,break,continue,range,goto,switch,case,fallthrough,if,else,default,},
206        morekeywords=[5]{Println,Printf,Error,},
207        sensitive=true,
208        morecomment=[l]{//},
209        morecomment=[s]{/*}{*/},
210        morestring=[b]',
211        morestring=[b]",
212        morestring=[s]{`}{`},
213}
214
215% Go programming language: https://github.com/julienc91/listings-golang/blob/master/listings-golang.sty
216\lstdefinelanguage{Golang}{
217        morekeywords=[1]{package,import,func,type,struct,return,defer,panic,recover,select,var,const,iota,},
218        morekeywords=[2]{string,uint,uint8,uint16,uint32,uint64,int,int8,int16,int32,int64,
219                bool,float32,float64,complex64,complex128,byte,rune,uintptr, error,interface},
220        morekeywords=[3]{map,slice,make,new,nil,len,cap,copy,close,true,false,delete,append,real,imag,complex,chan,},
221        morekeywords=[4]{for,break,continue,range,goto,switch,case,fallthrough,if,else,default,},
222        morekeywords=[5]{Println,Printf,Error,},
223        sensitive=true,
224        morecomment=[l]{//},
225        morecomment=[s]{/*}{*/},
226        morestring=[b]',
227        morestring=[b]",
228        morestring=[s]{`}{`},
229        % replace/adjust listing characters that look bad in sanserif
230        literate={-}{\makebox[1ex][c]{\raisebox{0.4ex}{\rule{0.8ex}{0.1ex}}}}1 {^}{\raisebox{0.6ex}{$\scriptstyle\land\,$}}1
231                {~}{\raisebox{0.3ex}{$\scriptstyle\sim\,$}}1 % {`}{\ttfamily\upshape\hspace*{-0.1ex}`}1
232                {<}{\textrm{\textless}}1 {>}{\textrm{\textgreater}}1
233                {<-}{\makebox[2ex][c]{\textrm{\textless}\raisebox{0.5ex}{\rule{0.8ex}{0.075ex}}}}2,
234}
235
236\lstnewenvironment{cfa}[1][]
237{\lstset{#1}}
238{}
239\lstnewenvironment{C++}[1][]                            % use C++ style
240{\lstset{language=C++,moredelim=**[is][\protect\color{red}]{`}{`},#1}\lstset{#1}}
241{}
242\lstnewenvironment{uC++}[1][]
243{\lstset{#1}}
244{}
245\lstnewenvironment{Go}[1][]
246{\lstset{language=Golang,moredelim=**[is][\protect\color{red}]{`}{`},#1}\lstset{#1}}
247{}
248\lstnewenvironment{python}[1][]
249{\lstset{language=python,moredelim=**[is][\protect\color{red}]{`}{`},#1}\lstset{#1}}
250{}
251
252% inline code @...@
253\lstMakeShortInline@%
254
255\let\OLDthebibliography\thebibliography
256\renewcommand\thebibliography[1]{
257  \OLDthebibliography{#1}
258  \setlength{\parskip}{0pt}
259  \setlength{\itemsep}{4pt plus 0.3ex}
260}
261
262\newbox\myboxA
263\newbox\myboxB
264\newbox\myboxC
265\newbox\myboxD
266
267\title{\texorpdfstring{Advanced Control-flow and Concurrency in \protect\CFA}{Advanced Control-flow in Cforall}}
268
269\author[1]{Thierry Delisle}
270\author[1]{Peter A. Buhr*}
271\authormark{DELISLE \textsc{et al.}}
272
273\address[1]{\orgdiv{Cheriton School of Computer Science}, \orgname{University of Waterloo}, \orgaddress{\state{Waterloo, ON}, \country{Canada}}}
274
275\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}}
276
277% \fundingInfo{Natural Sciences and Engineering Research Council of Canada}
278
279\abstract[Summary]{
280\CFA is a polymorphic, non-object-oriented, concurrent, backwards-compatible extension of the C programming language.
281This paper discusses the design philosophy and implementation of its advanced control-flow and concurrent/parallel features, along with the supporting runtime written in \CFA.
282These 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.
283\CFA introduces modern language-level control-flow mechanisms, like generators, coroutines, user-level threading, and monitors for mutual exclusion and synchronization.
284% Library extension for executors, futures, and actors are built on these basic mechanisms.
285The runtime provides significant programmer simplification and safety by eliminating spurious wakeup and monitor barging.
286The runtime also ensures multiple monitors can be safely acquired \emph{simultaneously} (deadlock free), and this feature is fully integrated with all monitor synchronization mechanisms.
287All control-flow features integrate with the \CFA polymorphic type-system and exception handling, while respecting the expectations and style of C programmers.
288Experimental results show comparable performance of the new features with similar mechanisms in other concurrent programming languages.
289}%
290
291\keywords{generator, coroutine, concurrency, parallelism, thread, monitor, runtime, C, \CFA (Cforall)}
292
293
294\begin{document}
295\linenumbers                                            % comment out to turn off line numbering
296
297\maketitle
298
299
300\section{Introduction}
301
302This 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.
303\CFA is a modern, polymorphic, non-object-oriented\footnote{
304\CFA has features often associated with object-oriented programming languages, such as constructors, destructors, virtuals and simple inheritance.
305However, 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.},
306backwards-compatible extension of the C programming language.
307In 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.
308Within 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}.
309However, \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;
310no high-level language concurrency features are defined.
311Interestingly, 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.
312Finally, 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}.
313
314In contrast, there has been a renewed interest during the past decade in user-level (M:N, green) threading in old and new programming languages.
315As multi-core hardware became available in the 1980/90s, both user and kernel threading were examined.
316Kernel 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}.
317Libraries like pthreads were developed for C, and the Solaris operating-system switched from user (JDK 1.1~\cite{JDK1.1}) to kernel threads.
318As 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.
319From 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}.
320The main argument for user-level threading is that it is lighter weight than kernel threading (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}.
321As well, user-threading facilitates a simpler concurrency approach using thread objects that leverage sequential patterns versus events with call-backs~\cite{Adya02,vonBehren03}.
322Finally, 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.
323
324A 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}.
325The 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.
326One solution is low-level qualifiers and functions (\eg @volatile@ and atomics) allowing \emph{programmers} to explicitly write safe (race-free~\cite{Boehm12}) programs.
327A safer solution is high-level language constructs so the \emph{compiler} knows the optimization boundaries, and hence, provides implicit safety.
328This problem is best known with respect to concurrency, but applies to other complex control-flow, like exceptions\footnote{
329\CFA exception handling will be presented in a separate paper.
330The 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++}
331} and coroutines.
332Finally, 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.
333
334Finally, it is important for a language to provide safety over performance \emph{as the default}, allowing careful reduction of safety for performance when necessary.
335Two 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.
336However, spurious wakeup is \emph{not} a foundational concurrency property~\cite[\S~8]{Buhr05a}, it is a performance design choice.
337Similarly, signals-as-hints are often a performance decision.
338We argue removing spurious wakeup and signals-as-hints make concurrent programming significantly safer because it removes local non-determinism and matches with programmer expectation.
339(Author experience teaching concurrency is that students are highly confused by these semantics.)
340Clawing back performance, when local non-determinism is unimportant, should be an option not the default.
341
342\begin{comment}
343Most 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.
344As a result, there is a significant learning curve to move to these languages, and C legacy-code must be rewritten.
345While \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.
346Hence, rewriting and retraining costs for these languages, even \CC, are prohibitive for companies with a large C software-base.
347\CFA with its orthogonal feature-set, its high-performance runtime, and direct access to all existing C libraries circumvents these problems.
348\end{comment}
349
350\CFA embraces user-level threading, language extensions for advanced control-flow, and safety as the default.
351We 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.
352The main contributions of this work are:
353\begin{itemize}
354\item
355language-level generators, coroutines and user-level threading, which respect the expectations of C programmers.
356\item
357monitor 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.
358\item
359providing statically type-safe interfaces that integrate with the \CFA polymorphic type-system and other language features.
360% \item
361% library extensions for executors, futures, and actors built on the basic mechanisms.
362\item
363a runtime system with no spurious wakeup.
364\item
365a dynamic partitioning mechanism to segregate the execution environment for specialized requirements.
366% \item
367% a non-blocking I/O library
368\item
369experimental results showing comparable performance of the new features with similar mechanisms in other programming languages.
370\end{itemize}
371
372
373\section{Stateful Function}
374
375The 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.
376Hence, 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.
377This capability is accomplished by retaining a data/execution \emph{closure} between invocations.
378If 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.
379If the closure is variable size, we call it a \emph{coroutine} (or \emph{stackful}), and as the names implies, often implemented with a separate stack with no programming restrictions.
380Hence, refactoring a stackless coroutine may require changing it to stackful.
381A 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.
382As well, activating a stateful function is \emph{asymmetric} or \emph{symmetric}, identified by resume/suspend (no cycles) and resume/resume (cycles).
383A fixed closure activated by modified call/return is faster than a variable closure activated by context switching.
384Additionally, any storage management for the closure (especially in unmanaged languages, \ie no garbage collection) must also be factored into design and performance.
385Therefore, selecting between stackless and stackful semantics is a tradeoff between programming requirements and performance, where stackless is faster and stackful is more general.
386Note, creation cost is amortized across usage, so activation cost is usually the dominant factor.
387
388\begin{figure}
389\centering
390\begin{lrbox}{\myboxA}
391\begin{cfa}[aboveskip=0pt,belowskip=0pt]
392typedef struct {
393        int fn1, fn;
394} Fib;
395#define FibCtor { 1, 0 }
396int fib( Fib * f ) {
397
398
399
400        int fn = f->fn; f->fn = f->fn1;
401                f->fn1 = f->fn + fn;
402        return fn;
403
404}
405int main() {
406        Fib f1 = FibCtor, f2 = FibCtor;
407        for ( int i = 0; i < 10; i += 1 )
408                printf( "%d %d\n",
409                           fib( &f1 ), fib( &f2 ) );
410}
411\end{cfa}
412\end{lrbox}
413
414\begin{lrbox}{\myboxB}
415\begin{cfa}[aboveskip=0pt,belowskip=0pt]
416`generator` Fib {
417        int fn1, fn;
418};
419
420void `main(Fib & fib)` with(fib) {
421
422        [fn1, fn] = [1, 0];
423        for () {
424                `suspend;`
425                [fn1, fn] = [fn, fn + fn1];
426
427        }
428}
429int main() {
430        Fib f1, f2;
431        for ( 10 )
432                sout | `resume( f1 )`.fn
433                         | `resume( f2 )`.fn;
434}
435\end{cfa}
436\end{lrbox}
437
438\begin{lrbox}{\myboxC}
439\begin{cfa}[aboveskip=0pt,belowskip=0pt]
440typedef struct {
441        int fn1, fn;  void * `next`;
442} Fib;
443#define FibCtor { 1, 0, NULL }
444Fib * comain( Fib * f ) {
445        if ( f->next ) goto *f->next;
446        f->next = &&s1;
447        for ( ;; ) {
448                return f;
449          s1:; int fn = f->fn + f->fn1;
450                        f->fn1 = f->fn; f->fn = fn;
451        }
452}
453int main() {
454        Fib f1 = FibCtor, f2 = FibCtor;
455        for ( int i = 0; i < 10; i += 1 )
456                printf("%d %d\n",comain(&f1)->fn,
457                                 comain(&f2)->fn);
458}
459\end{cfa}
460\end{lrbox}
461
462\subfloat[C asymmetric generator]{\label{f:CFibonacci}\usebox\myboxA}
463\hspace{3pt}
464\vrule
465\hspace{3pt}
466\subfloat[\CFA asymmetric generator]{\label{f:CFAFibonacciGen}\usebox\myboxB}
467\hspace{3pt}
468\vrule
469\hspace{3pt}
470\subfloat[C generator implementation]{\label{f:CFibonacciSim}\usebox\myboxC}
471\caption{Fibonacci (output) asymmetric generator}
472\label{f:FibonacciAsymmetricGenerator}
473
474\bigskip
475
476\begin{lrbox}{\myboxA}
477\begin{cfa}[aboveskip=0pt,belowskip=0pt]
478`generator Fmt` {
479        char ch;
480        int g, b;
481};
482void ?{}( Fmt & fmt ) { `resume(fmt);` } // constructor
483void ^?{}( Fmt & f ) with(f) { $\C[1.75in]{// destructor}$
484        if ( g != 0 || b != 0 ) sout | nl; }
485void `main( Fmt & f )` with(f) {
486        for () { $\C{// until destructor call}$
487                for ( ; g < 5; g += 1 ) { $\C{// groups}$
488                        for ( ; b < 4; b += 1 ) { $\C{// blocks}$
489                                `suspend;` $\C{// wait for character}$
490                                while ( ch == '\n' ) `suspend;` // ignore
491                                sout | ch;                                              // newline
492                        } sout | " ";  // block spacer
493                } sout | nl; // group newline
494        }
495}
496int main() {
497        Fmt fmt; $\C{// fmt constructor called}$
498        for () {
499                sin | fmt.ch; $\C{// read into generator}$
500          if ( eof( sin ) ) break;
501                `resume( fmt );`
502        }
503
504} $\C{// fmt destructor called}\CRT$
505\end{cfa}
506\end{lrbox}
507
508\begin{lrbox}{\myboxB}
509\begin{cfa}[aboveskip=0pt,belowskip=0pt]
510typedef struct {
511        void * next;
512        char ch;
513        int g, b;
514} Fmt;
515void comain( Fmt * f ) {
516        if ( f->next ) goto *f->next;
517        f->next = &&s1;
518        for ( ;; ) {
519                for ( f->g = 0; f->g < 5; f->g += 1 ) {
520                        for ( f->b = 0; f->b < 4; f->b += 1 ) {
521                                return;
522                          s1:;  while ( f->ch == '\n' ) return;
523                                printf( "%c", f->ch );
524                        } printf( " " );
525                } printf( "\n" );
526        }
527}
528int main() {
529        Fmt fmt = { NULL };  comain( &fmt ); // prime
530        for ( ;; ) {
531                scanf( "%c", &fmt.ch );
532          if ( feof( stdin ) ) break;
533                comain( &fmt );
534        }
535        if ( fmt.g != 0 || fmt.b != 0 ) printf( "\n" );
536}
537\end{cfa}
538\end{lrbox}
539
540\subfloat[\CFA asymmetric generator]{\label{f:CFAFormatGen}\usebox\myboxA}
541\hspace{3pt}
542\vrule
543\hspace{3pt}
544\subfloat[C generator simulation]{\label{f:CFormatSim}\usebox\myboxB}
545\hspace{3pt}
546\caption{Formatter (input) asymmetric generator}
547\label{f:FormatterAsymmetricGenerator}
548\end{figure}
549
550Stateful functions appear as generators, coroutines, and threads, where presentations are based on function objects or pointers~\cite{Butenhof97, C++14, MS:VisualC++, BoostCoroutines15}.
551For example, Python presents generators as a function object:
552\begin{python}
553def Gen():
554        ... `yield val` ...
555gen = Gen()
556for i in range( 10 ):
557        print( next( gen ) )
558\end{python}
559Boost presents coroutines in terms of four functor object-types:
560\begin{cfa}
561asymmetric_coroutine<>::pull_type
562asymmetric_coroutine<>::push_type
563symmetric_coroutine<>::call_type
564symmetric_coroutine<>::yield_type
565\end{cfa}
566and many languages present threading using function pointers, @pthreads@~\cite{Butenhof97}, \Csharp~\cite{Csharp}, Go~\cite{Go}, and Scala~\cite{Scala}, \eg pthreads:
567\begin{cfa}
568void * rtn( void * arg ) { ... }
569int i = 3, rc;
570pthread_t t; $\C{// thread id}$
571`rc = pthread_create( &t, rtn, (void *)i );` $\C{// create and initialized task, type-unsafe input parameter}$
572\end{cfa}
573% void mycor( pthread_t cid, void * arg ) {
574%       int * value = (int *)arg;                               $\C{// type unsafe, pointer-size only}$
575%       // thread body
576% }
577% int main() {
578%       int input = 0, output;
579%       coroutine_t cid = coroutine_create( &mycor, (void *)&input ); $\C{// type unsafe, pointer-size only}$
580%       coroutine_resume( cid, (void *)input, (void **)&output ); $\C{// type unsafe, pointer-size only}$
581% }
582\CFA's preferred presentation model for generators/coroutines/threads is a hybrid of objects and functions, with an object-oriented flavour.
583Essentially, the generator/coroutine/thread function is semantically coupled with a generator/coroutine/thread custom type.
584The custom type solves several issues, while accessing the underlying mechanisms used by the custom types is still allowed.
585
586
587\subsection{Generator}
588
589Stackless generators have the potential to be very small and fast, \ie as small and fast as function call/return for both creation and execution.
590The \CFA goal is to achieve this performance target, possibly at the cost of some semantic complexity.
591A series of different kinds of generators and their implementation demonstrate how this goal is accomplished.
592
593Figure~\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.
594This generator is an \emph{output generator}, producing a new result on each resumption.
595To 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.
596An additional requirement is the ability to create an arbitrary number of generators (of any kind), \ie retaining one state in global variables is insufficient;
597hence, state is retained in a closure between calls.
598Figure~\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.
599The C version only has the middle execution state because the top execution state is declaration initialization.
600Figure~\ref{f:CFAFibonacciGen} shows the \CFA approach, which also has a manual closure, but replaces the structure with a custom \CFA @generator@ type.
601This generator type is then connected to a function that \emph{must be named \lstinline|main|},\footnote{
602The name \lstinline|main| has special meaning in C, specifically the function where a program starts execution.
603Hence, overloading this name for other starting points (generator/coroutine/thread) is a logical extension.}
604called a \emph{generator main},which takes as its only parameter a reference to the generator type.
605The generator main contains @suspend@ statements that suspend execution without ending the generator versus @return@.
606For the Fibonacci generator-main,\footnote{
607The \CFA \lstinline|with| opens an aggregate scope making its fields directly accessible, like Pascal \lstinline|with|, but using parallel semantics.
608Multiple aggregates may be opened.}
609the top initialization state appears at the start and the middle execution state is denoted by statement @suspend@.
610Any local variables in @main@ \emph{are not retained} between calls;
611hence local variables are only for temporary computations \emph{between} suspends.
612All retained state \emph{must} appear in the generator's type.
613As 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.
614The 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.
615Resuming an ended (returned) generator is undefined.
616Function @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.
617Figure~\ref{f:CFibonacciSim} shows the C implementation of the \CFA generator only needs one additional field, @next@, to handle retention of execution state.
618The computed @goto@ at the start of the generator main, which branches after the previous suspend, adds very little cost to the resume call.
619Finally, an explicit generator type provides both design and performance benefits, such as multiple type-safe interface functions taking and returning arbitrary types.
620\begin{cfa}
621int ?()( Fib & fib ) { return `resume( fib )`.fn; } $\C[3.9in]{// function-call interface}$
622int ?()( Fib & fib, int N ) { for ( N - 1 ) `fib()`; return `fib()`; } $\C{// use function-call interface to skip N values}$
623double ?()( Fib & fib ) { return (int)`fib()` / 3.14159; } $\C{// different return type, cast prevents recursive call}\CRT$
624sout | (int)f1() | (double)f1() | f2( 2 ); // alternative interface, cast selects call based on return type, step 2 values
625\end{cfa}
626Now, the generator can be a separately compiled opaque-type only accessed through its interface functions.
627For 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.
628
629Having to manually create the generator closure by moving local-state variables into the generator type is an additional programmer burden.
630(This restriction is removed by the coroutine in Section~\ref{s:Coroutine}.)
631This 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.
632However, dynamic allocation significantly increases the cost of generator creation/destruction and is a showstopper for embedded real-time programming.
633But 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.
634With 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.
635Finally, 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.
636As well, C programmers are not afraid of this kind of semantic programming requirement, if it results in very small, fast generators.
637
638Figure~\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.
639\begin{center}
640\tt
641\begin{tabular}{@{}l|l@{}}
642\multicolumn{1}{c|}{\textbf{\textrm{input}}} & \multicolumn{1}{c}{\textbf{\textrm{output}}} \\
643\begin{tabular}[t]{@{}ll@{}}
644abcdefghijklmnopqrstuvwxyz \\
645abcdefghijklmnopqrstuvwxyz
646\end{tabular}
647&
648\begin{tabular}[t]{@{}lllll@{}}
649abcd    & efgh  & ijkl  & mnop  & qrst  \\
650uvwx    & yzab  & cdef  & ghij  & klmn  \\
651opqr    & stuv  & wxyz  &               &
652\end{tabular}
653\end{tabular}
654\end{center}
655The 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.
656The destructor provides a newline, if formatted text ends with a full line.
657Figure~\ref{f:CFormatSim} shows the C implementation of the \CFA input generator with one additional field and the computed @goto@.
658For contrast, Figure~\ref{f:PythonFormatter} shows the equivalent Python format generator with the same properties as the Fibonacci generator.
659
660Figure~\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}.
661Device drives follow the pattern of simple data state but complex execution state, \ie finite state-machine (FSM) parsing a protocol.
662For example, the following protocol:
663\begin{center}
664\ldots\, STX \ldots\, message \ldots\, ESC ETX \ldots\, message \ldots\, ETX 2-byte crc \ldots
665\end{center}
666is a network message beginning with the control character STX, ending with an ETX, and followed by a 2-byte cyclic-redundancy check.
667Control characters may appear in a message if preceded by an ESC.
668When 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.
669The 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.
670Hence, the device driver is an input/output generator.
671
672Note, 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.
673As 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.
674% Manually, detecting and hoisting local-state variables is easy when the number is small.
675In contrast, the execution state is large, with one @resume@ and seven @suspend@s.
676Hence, 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.
677Because FSMs can be complex and frequently occur in important domains, direct generator support is important in a system programming language.
678
679\begin{figure}
680\centering
681\newbox\myboxA
682\begin{lrbox}{\myboxA}
683\begin{python}[aboveskip=0pt,belowskip=0pt]
684def Fib():
685        fn1, fn = 0, 1
686        while True:
687                `yield fn1`
688                fn1, fn = fn, fn1 + fn
689f1 = Fib()
690f2 = Fib()
691for i in range( 10 ):
692        print( next( f1 ), next( f2 ) )
693
694
695
696
697
698
699\end{python}
700\end{lrbox}
701
702\newbox\myboxB
703\begin{lrbox}{\myboxB}
704\begin{python}[aboveskip=0pt,belowskip=0pt]
705def Fmt():
706        try:
707                while True:
708                        for g in range( 5 ):
709                                for b in range( 4 ):
710                                        print( `(yield)`, end='' )
711                                print( '  ', end='' )
712                        print()
713        except GeneratorExit:
714                if g != 0 | b != 0:
715                        print()
716fmt = Fmt()
717`next( fmt )`                    # prime, next prewritten
718for i in range( 41 ):
719        `fmt.send( 'a' );`      # send to yield
720\end{python}
721\end{lrbox}
722\subfloat[Fibonacci]{\label{f:PythonFibonacci}\usebox\myboxA}
723\hspace{3pt}
724\vrule
725\hspace{3pt}
726\subfloat[Formatter]{\label{f:PythonFormatter}\usebox\myboxB}
727\caption{Python generator}
728\label{f:PythonGenerator}
729
730\bigskip
731
732\begin{tabular}{@{}l|l@{}}
733\begin{cfa}[aboveskip=0pt,belowskip=0pt]
734enum Status { CONT, MSG, ESTX,
735                                ELNTH, ECRC };
736`generator` Driver {
737        Status status;
738        unsigned char byte, * msg; // communication
739        unsigned int lnth, sum;      // local state
740        unsigned short int crc;
741};
742void ?{}( Driver & d, char * m ) { d.msg = m; }
743Status next( Driver & d, char b ) with( d ) {
744        byte = b; `resume( d );` return status;
745}
746void main( Driver & d ) with( d ) {
747        enum { STX = '\002', ESC = '\033',
748                        ETX = '\003', MaxMsg = 64 };
749  msg: for () { // parse message
750                status = CONT;
751                lnth = 0; sum = 0;
752                while ( byte != STX ) `suspend;`
753          emsg: for () {
754                        `suspend;` // process byte
755\end{cfa}
756&
757\begin{cfa}[aboveskip=0pt,belowskip=0pt]
758                        choose ( byte ) { // switch with implicit break
759                          case STX:
760                                status = ESTX; `suspend;` continue msg;
761                          case ETX:
762                                break emsg;
763                          case ESC:
764                                `suspend;`
765                        }
766                        if ( lnth >= MaxMsg ) { // buffer full ?
767                                status = ELNTH; `suspend;` continue msg; }
768                        msg[lnth++] = byte;
769                        sum += byte;
770                }
771                msg[lnth] = '\0'; // terminate string
772                `suspend;`
773                crc = byte << 8;
774                `suspend;`
775                status = (crc | byte) == sum ? MSG : ECRC;
776                `suspend;`
777        }
778}
779\end{cfa}
780\end{tabular}
781\caption{Device-driver generator for communication protocol}
782\label{f:DeviceDriverGen}
783\end{figure}
784
785Figure~\ref{f:CFAPingPongGen} shows a symmetric generator, where the generator resumes another generator, forming a resume/resume cycle.
786(The trivial cycle is a generator resuming itself.)
787This control flow is similar to recursion for functions but without stack growth.
788The steps for symmetric control-flow are creating, executing, and terminating the cycle.
789Constructing 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.
790(This issue occurs for any cyclic data structure.)
791% The example creates all the generators and then assigns the partners that form the cycle.
792% 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.
793Once 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).
794Terminating 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).
795The starting stack-frame is below the last active generator because the resume/resume cycle does not grow the stack.
796Also, 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.
797Destructor cost occurs when the generator instance is deallocated, which is easily controlled by the programmer.
798
799Figure~\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.
800This 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.
801However, before the jump, the caller must reset its stack (and any registers) equivalent to a @return@, but subsequently jump forward.
802This semantics is basically a tail-call optimization, which compilers already perform.
803The example shows the assembly code to undo the generator's entry code before the direct jump.
804This assembly code depends on what entry code is generated, specifically if there are local variables and the level of optimization.
805To provide this new calling convention requires a mechanism built into the compiler, which is beyond the scope of \CFA at this time.
806Nevertheless, it is possible to hand generate any symmetric generators for proof of concept and performance testing.
807A 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@.
808
809\begin{figure}
810\centering
811\begin{lrbox}{\myboxA}
812\begin{cfa}[aboveskip=0pt,belowskip=0pt]
813`generator PingPong` {
814        const char * name;
815        int N;
816        int i;                          // local state
817        PingPong & partner; // rebindable reference
818};
819
820void `main( PingPong & pp )` with(pp) {
821        for ( ; i < N; i += 1 ) {
822                sout | name | i;
823                `resume( partner );`
824        }
825}
826int main() {
827        enum { N = 5 };
828        PingPong ping = {"ping",N,0}, pong = {"pong",N,0};
829        &ping.partner = &pong;  &pong.partner = &ping;
830        `resume( ping );`
831}
832\end{cfa}
833\end{lrbox}
834
835\begin{lrbox}{\myboxB}
836\begin{cfa}[escapechar={},aboveskip=0pt,belowskip=0pt]
837typedef struct PingPong {
838        const char * name;
839        int N, i;
840        struct PingPong * partner;
841        void * next;
842} PingPong;
843#define PPCtor(name, N) {name,N,0,NULL,NULL}
844void comain( PingPong * pp ) {
845        if ( pp->next ) goto *pp->next;
846        pp->next = &&cycle;
847        for ( ; pp->i < pp->N; pp->i += 1 ) {
848                printf( "%s %d\n", pp->name, pp->i );
849                asm( "mov  %0,%%rdi" : "=m" (pp->partner) );
850                asm( "mov  %rdi,%rax" );
851                asm( "popq %rbx" );
852                asm( "jmp  comain" );
853          cycle: ;
854        }
855}
856\end{cfa}
857\end{lrbox}
858
859\subfloat[\CFA symmetric generator]{\label{f:CFAPingPongGen}\usebox\myboxA}
860\hspace{3pt}
861\vrule
862\hspace{3pt}
863\subfloat[C generator simulation]{\label{f:CPingPongSim}\usebox\myboxB}
864\hspace{3pt}
865\caption{Ping-Pong symmetric generator}
866\label{f:PingPongSymmetricGenerator}
867\end{figure}
868
869Finally, part of this generator work was inspired by the recent \CCtwenty generator proposal~\cite{C++20Coroutine19} (which they call coroutines).
870Our work provides the same high-performance asymmetric generators as \CCtwenty, and extends their work with symmetric generators.
871An 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:
872\begin{cfa}
873... suspend`{ ... }`;
874... resume( C )`{ ... }` ...
875\end{cfa}
876Since the current generator's stack is released before calling the compound statement, the compound statement can only reference variables in the generator's type.
877This 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.
878Hence, this mechanism provides a general and safe handoff of the generator among competing threads.
879
880
881\subsection{Coroutine}
882\label{s:Coroutine}
883
884Stackful coroutines extend generator semantics, \ie there is an implicit closure and @suspend@ may appear in a helper function called from the coroutine main.
885A coroutine is specified by replacing @generator@ with @coroutine@ for the type.
886Coroutine 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.
887A series of different kinds of coroutines and their implementations demonstrate how coroutines extend generators.
888
889First, 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.
890\begin{description}
891\item[Fibonacci]
892Move the declaration of @fn1@ to the start of coroutine main.
893\begin{cfa}[xleftmargin=0pt]
894void main( Fib & fib ) with(fib) {
895        `int fn1;`
896\end{cfa}
897\item[Formatter]
898Move the declaration of @g@ and @b@ to the for loops in the coroutine main.
899\begin{cfa}[xleftmargin=0pt]
900for ( `g`; 5 ) {
901        for ( `b`; 4 ) {
902\end{cfa}
903\item[Device Driver]
904Move the declaration of @lnth@ and @sum@ to their points of initialization.
905\begin{cfa}[xleftmargin=0pt]
906        status = CONT;
907        `unsigned int lnth = 0, sum = 0;`
908        ...
909        `unsigned short int crc = byte << 8;`
910\end{cfa}
911\item[PingPong]
912Move the declaration of @i@ to the for loop in the coroutine main.
913\begin{cfa}[xleftmargin=0pt]
914void main( PingPong & pp ) with(pp) {
915        for ( `i`; N ) {
916\end{cfa}
917\end{description}
918It 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.
919\begin{cfa}
920unsigned int Crc() {
921        `suspend;`
922        unsigned short int crc = byte << 8;
923        `suspend;`
924        status = (crc | byte) == sum ? MSG : ECRC;
925        return crc;
926}
927\end{cfa}
928A call to this function is placed at the end of the driver's coroutine-main.
929For complex finite-state machines, refactoring is part of normal program abstraction, especially when code is used in multiple places.
930Again, this complexity is usually associated with execution state rather than data state.
931
932\begin{comment}
933Figure~\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@.
934Like 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.
935The 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@.
936The interface function @next@, takes a Fibonacci instance and context switches to it using @resume@;
937on restart, the Fibonacci field, @fn@, contains the next value in the sequence, which is returned.
938The first @resume@ is special because it allocates the coroutine stack and cocalls its coroutine main on that stack;
939when the coroutine main returns, its stack is deallocated.
940Hence, @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.
941Figure~\ref{f:Coroutine1State} shows the coroutine version of the C version in Figure~\ref{f:ExternalState}.
942Coroutine generators are called \newterm{output coroutines} because values are only returned.
943
944\begin{figure}
945\centering
946\newbox\myboxA
947% \begin{lrbox}{\myboxA}
948% \begin{cfa}[aboveskip=0pt,belowskip=0pt]
949% `int fn1, fn2, state = 1;`   // single global variables
950% int fib() {
951%       int fn;
952%       `switch ( state )` {  // explicit execution state
953%         case 1: fn = 0;  fn1 = fn;  state = 2;  break;
954%         case 2: fn = 1;  fn2 = fn1;  fn1 = fn;  state = 3;  break;
955%         case 3: fn = fn1 + fn2;  fn2 = fn1;  fn1 = fn;  break;
956%       }
957%       return fn;
958% }
959% int main() {
960%
961%       for ( int i = 0; i < 10; i += 1 ) {
962%               printf( "%d\n", fib() );
963%       }
964% }
965% \end{cfa}
966% \end{lrbox}
967\begin{lrbox}{\myboxA}
968\begin{cfa}[aboveskip=0pt,belowskip=0pt]
969#define FibCtor { 0, 1 }
970typedef struct { int fn1, fn; } Fib;
971int fib( Fib * f ) {
972
973        int ret = f->fn1;
974        f->fn1 = f->fn;
975        f->fn = ret + f->fn;
976        return ret;
977}
978
979
980
981int main() {
982        Fib f1 = FibCtor, f2 = FibCtor;
983        for ( int i = 0; i < 10; i += 1 ) {
984                printf( "%d %d\n",
985                                fib( &f1 ), fib( &f2 ) );
986        }
987}
988\end{cfa}
989\end{lrbox}
990
991\newbox\myboxB
992\begin{lrbox}{\myboxB}
993\begin{cfa}[aboveskip=0pt,belowskip=0pt]
994`coroutine` Fib { int fn1; };
995void main( Fib & fib ) with( fib ) {
996        int fn;
997        [fn1, fn] = [0, 1];
998        for () {
999                `suspend;`
1000                [fn1, fn] = [fn, fn1 + fn];
1001        }
1002}
1003int ?()( Fib & fib ) with( fib ) {
1004        return `resume( fib )`.fn1;
1005}
1006int main() {
1007        Fib f1, f2;
1008        for ( 10 ) {
1009                sout | f1() | f2();
1010}
1011
1012
1013\end{cfa}
1014\end{lrbox}
1015
1016\newbox\myboxC
1017\begin{lrbox}{\myboxC}
1018\begin{python}[aboveskip=0pt,belowskip=0pt]
1019
1020def Fib():
1021
1022        fn1, fn = 0, 1
1023        while True:
1024                `yield fn1`
1025                fn1, fn = fn, fn1 + fn
1026
1027
1028// next prewritten
1029
1030
1031f1 = Fib()
1032f2 = Fib()
1033for i in range( 10 ):
1034        print( next( f1 ), next( f2 ) )
1035
1036
1037
1038\end{python}
1039\end{lrbox}
1040
1041\subfloat[C]{\label{f:GlobalVariables}\usebox\myboxA}
1042\hspace{3pt}
1043\vrule
1044\hspace{3pt}
1045\subfloat[\CFA]{\label{f:ExternalState}\usebox\myboxB}
1046\hspace{3pt}
1047\vrule
1048\hspace{3pt}
1049\subfloat[Python]{\label{f:ExternalState}\usebox\myboxC}
1050\caption{Fibonacci generator}
1051\label{f:C-fibonacci}
1052\end{figure}
1053
1054\bigskip
1055
1056\newbox\myboxA
1057\begin{lrbox}{\myboxA}
1058\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1059`coroutine` Fib { int fn; };
1060void main( Fib & fib ) with( fib ) {
1061        fn = 0;  int fn1 = fn; `suspend`;
1062        fn = 1;  int fn2 = fn1;  fn1 = fn; `suspend`;
1063        for () {
1064                fn = fn1 + fn2; fn2 = fn1; fn1 = fn; `suspend`; }
1065}
1066int next( Fib & fib ) with( fib ) { `resume( fib );` return fn; }
1067int main() {
1068        Fib f1, f2;
1069        for ( 10 )
1070                sout | next( f1 ) | next( f2 );
1071}
1072\end{cfa}
1073\end{lrbox}
1074\newbox\myboxB
1075\begin{lrbox}{\myboxB}
1076\begin{python}[aboveskip=0pt,belowskip=0pt]
1077
1078def Fibonacci():
1079        fn = 0; fn1 = fn; `yield fn`  # suspend
1080        fn = 1; fn2 = fn1; fn1 = fn; `yield fn`
1081        while True:
1082                fn = fn1 + fn2; fn2 = fn1; fn1 = fn; `yield fn`
1083
1084
1085f1 = Fibonacci()
1086f2 = Fibonacci()
1087for i in range( 10 ):
1088        print( `next( f1 )`, `next( f2 )` ) # resume
1089
1090\end{python}
1091\end{lrbox}
1092\subfloat[\CFA]{\label{f:Coroutine3States}\usebox\myboxA}
1093\qquad
1094\subfloat[Python]{\label{f:Coroutine1State}\usebox\myboxB}
1095\caption{Fibonacci input coroutine, 3 states, internal variables}
1096\label{f:cfa-fibonacci}
1097\end{figure}
1098\end{comment}
1099
1100\begin{figure}
1101\centering
1102\lstset{language=CFA,escapechar={},moredelim=**[is][\protect\color{red}]{`}{`}}% allow $
1103\begin{tabular}{@{}l@{\hspace{2\parindentlnth}}l@{}}
1104\begin{cfa}
1105`coroutine` Prod {
1106        Cons & c;                       // communication
1107        int N, money, receipt;
1108};
1109void main( Prod & prod ) with( prod ) {
1110        // 1st resume starts here
1111        for ( i; N ) {
1112                int p1 = random( 100 ), p2 = random( 100 );
1113                sout | p1 | " " | p2;
1114                int status = delivery( c, p1, p2 );
1115                sout | " $" | money | nl | status;
1116                receipt += 1;
1117        }
1118        stop( c );
1119        sout | "prod stops";
1120}
1121int payment( Prod & prod, int money ) {
1122        prod.money = money;
1123        `resume( prod );`
1124        return prod.receipt;
1125}
1126void start( Prod & prod, int N, Cons &c ) {
1127        &prod.c = &c;
1128        prod.[N, receipt] = [N, 0];
1129        `resume( prod );`
1130}
1131int main() {
1132        Prod prod;
1133        Cons cons = { prod };
1134        start( prod, 5, cons );
1135}
1136\end{cfa}
1137&
1138\begin{cfa}
1139`coroutine` Cons {
1140        Prod & p;                       // communication
1141        int p1, p2, status;
1142        bool done;
1143};
1144void ?{}( Cons & cons, Prod & p ) {
1145        &cons.p = &p; // reassignable reference
1146        cons.[status, done ] = [0, false];
1147}
1148void main( Cons & cons ) with( cons ) {
1149        // 1st resume starts here
1150        int money = 1, receipt;
1151        for ( ; ! done; ) {
1152                sout | p1 | " " | p2 | nl | " $" | money;
1153                status += 1;
1154                receipt = payment( p, money );
1155                sout | " #" | receipt;
1156                money += 1;
1157        }
1158        sout | "cons stops";
1159}
1160int delivery( Cons & cons, int p1, int p2 ) {
1161        cons.[p1, p2] = [p1, p2];
1162        `resume( cons );`
1163        return cons.status;
1164}
1165void stop( Cons & cons ) {
1166        cons.done = true;
1167        `resume( cons );`
1168}
1169
1170\end{cfa}
1171\end{tabular}
1172\caption{Producer / consumer: resume-resume cycle, bidirectional communication}
1173\label{f:ProdCons}
1174\end{figure}
1175
1176Figure~\ref{f:ProdCons} shows the ping-pong example in Figure~\ref{f:CFAPingPongGen} extended into a producer/consumer symmetric-coroutine performing bidirectional communication.
1177This example is illustrative because both producer/consumer have two interface functions with @resume@s that suspend execution in these interface (helper) functions.
1178The 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.
1179The first @resume@ of @prod@ creates @prod@'s stack with a frame for @prod@'s coroutine main at the top, and context switches to it.
1180@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.
1181
1182The producer call to @delivery@ transfers values into the consumer's communication variables, resumes the consumer, and returns the consumer status.
1183On the first resume, @cons@'s stack is created and initialized, holding local-state variables retained between subsequent activations of the coroutine.
1184The 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).
1185The call from the consumer to @payment@ introduces the cycle between producer and consumer.
1186When @payment@ is called, the consumer copies values into the producer's communication variable and a resume is executed.
1187The context switch restarts the producer at the point where it last context switched, so it continues in @delivery@ after the resume.
1188@delivery@ returns the status value in @prod@'s coroutine main, where the status is printed.
1189The loop then repeats calling @delivery@, where each call resumes the consumer coroutine.
1190The context switch to the consumer continues in @payment@.
1191The consumer increments and returns the receipt to the call in @cons@'s coroutine main.
1192The loop then repeats calling @payment@, where each call resumes the producer coroutine.
1193Figure~\ref{f:ProdConsRuntimeStacks} shows the runtime stacks of the program main, and the coroutine mains for @prod@ and @cons@ during the cycling.
1194
1195\begin{figure}
1196\begin{center}
1197\input{FullProdConsStack.pstex_t}
1198\end{center}
1199\vspace*{-10pt}
1200\caption{Producer / consumer runtime stacks}
1201\label{f:ProdConsRuntimeStacks}
1202
1203\medskip
1204
1205\begin{center}
1206\input{FullCoroutinePhases.pstex_t}
1207\end{center}
1208\vspace*{-10pt}
1209\caption{Ping / Pong coroutine steps}
1210\label{f:PingPongFullCoroutineSteps}
1211\end{figure}
1212
1213Terminating 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.
1214Furthermore, 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.
1215When a coroutine's main ends, its stack is already unwound so any stack allocated objects with destructors have been finalized.
1216The na\"{i}ve semantics for coroutine-cycle termination is to context switch to the last resumer, like executing a @suspend@/@return@ in a generator.
1217However, for coroutines, the last resumer is \emph{not} implicitly below the current stack frame, as for generators, because each coroutine's stack is independent.
1218Unfortunately, it is impossible to determine statically if a coroutine is in a cycle and unrealistic to check dynamically (graph-cycle problem).
1219Hence, a compromise solution is necessary that works for asymmetric (acyclic) and symmetric (cyclic) coroutines.
1220
1221Our solution is to context switch back to the first resumer (starter) once the coroutine ends.
1222This semantics works well for the most common asymmetric and symmetric coroutine usage patterns.
1223For asymmetric coroutines, it is common for the first resumer (starter) coroutine to be the only resumer.
1224All previous generators converted to coroutines have this property.
1225For symmetric coroutines, it is common for the cycle creator to persist for the lifetime of the cycle.
1226Hence, the starter coroutine is remembered on the first resume and ending the coroutine resumes the starter.
1227Figure~\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.
1228For 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.
1229
1230The producer/consumer example does not illustrate the full power of the starter semantics because @cons@ always ends first.
1231Assume generator @PingPong@ is converted to a coroutine.
1232Figure~\ref{f:PingPongFullCoroutineSteps} shows the creation, starter, and cyclic execution steps of the coroutine version.
1233The program main creates (declares) coroutine instances @ping@ and @pong@.
1234Next, program main resumes @ping@, making it @ping@'s starter, and @ping@'s main resumes @pong@'s main, making it @pong@'s starter.
1235Execution forms a cycle when @pong@ resumes @ping@, and cycles $N$ times.
1236By adjusting $N$ for either @ping@/@pong@, it is possible to have either one finish first, instead of @pong@ always ending first.
1237If @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@.
1238If @ping@ ends first, it resumes its starter the program main in function @start@.
1239Regardless of the cycle complexity, the starter stack always leads back to the program main, but the stack can be entered at an arbitrary point.
1240Once back at the program main, coroutines @ping@ and @pong@ are deallocated.
1241For generators, deallocation runs the destructors for all objects in the generator type.
1242For 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.
1243Hence, 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.
1244So 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.
1245Explicitly 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.
1246
1247Finally, there is an interesting effect for @suspend@ with symmetric coroutines.
1248A coroutine must retain its last resumer to suspend back because the resumer is on a different stack.
1249These reverse pointers allow @suspend@ to cycle \emph{backwards}, which may be useful in certain cases.
1250However, 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.
1251To prevent losing this information, a self-resume does not overwrite the last resumer.
1252
1253
1254\subsection{Generator / Coroutine Implementation}
1255
1256A 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.
1257There are several solutions to these problem, which follow from the object-oriented flavour of adopting custom types.
1258
1259For object-oriented languages, inheritance is used to provide extra fields and code via explicit inheritance:
1260\begin{cfa}[morekeywords={class,inherits}]
1261class myCoroutine inherits baseCoroutine { ... }
1262\end{cfa}
1263% 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.
1264The 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.
1265Alternatives, such as explicitly starting threads as in Java, are repetitive and forgetting to call start is a common source of errors.
1266An alternative is composition:
1267\begin{cfa}
1268struct myCoroutine {
1269        ... // declaration/communication variables
1270        baseCoroutine dummy; // composition, last declaration
1271}
1272\end{cfa}
1273which also requires an explicit declaration that must be last to ensure correct initialization order.
1274However, there is nothing preventing wrong placement or multiple declarations.
1275
1276\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
1277% and when it is unsafe to perform certain optimizations,
1278and IDEs using simple parsing can find and manipulate types with special properties.
1279The downside of this approach is that it makes custom types a special case in the language.
1280Users wanting to extend custom types or build their own can only do so in ways offered by the language.
1281Furthermore, implementing custom types without language support may display the power of a programming language.
1282\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.
1283
1284Part 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.
1285\begin{cfa}
1286trait is_coroutine( `dtype` T ) {
1287        void main( T & );
1288        coroutine_desc * get_coroutine( T & );
1289};
1290forall( `dtype` T | is_coroutine(T) ) void $suspend$( T & ), resume( T & );
1291\end{cfa}
1292Note, copying generators/coroutines/threads is not meaningful.
1293For example, both the resumer and suspender descriptors can have bidirectional pointers;
1294copying these coroutines does not update the internal pointers so behaviour of both copies would be difficult to understand.
1295Furthermore, two coroutines cannot logically execute on the same stack.
1296A 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.
1297The \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).
1298The 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.
1299The @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.
1300The 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@.
1301
1302The \CFA custom-type @coroutine@ implicitly implements the getter and forward declarations for the coroutine main.
1303\begin{cquote}
1304\begin{tabular}{@{}ccc@{}}
1305\begin{cfa}
1306coroutine MyCor {
1307        int value;
1308
1309};
1310\end{cfa}
1311&
1312{\Large $\Rightarrow$}
1313&
1314\begin{tabular}{@{}ccc@{}}
1315\begin{cfa}
1316struct MyCor {
1317        int value;
1318        coroutine_desc cor;
1319};
1320\end{cfa}
1321&
1322\begin{cfa}
1323static inline coroutine_desc *
1324get_coroutine( MyCor & this ) {
1325        return &this.cor;
1326}
1327\end{cfa}
1328&
1329\begin{cfa}
1330void main( MyCor * this );
1331
1332
1333
1334\end{cfa}
1335\end{tabular}
1336\end{tabular}
1337\end{cquote}
1338The 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.
1339
1340Figure~\ref{f:CoroutineMemoryLayout} shows different memory-layout options for a coroutine (where a task is similar).
1341The coroutine handle is the @coroutine@ instance containing programmer specified type global/communication variables across interface functions.
1342The coroutine descriptor contains all implicit declarations needed by the runtime, \eg @suspend@/@resume@, and can be part of the coroutine handle or separate.
1343The coroutine stack can appear in a number of locations and be fixed or variable sized.
1344Hence, the coroutine's stack could be a VLS\footnote{
1345We are examining variable-sized structures (VLS), where fields can be variable-sized structures or arrays.
1346Once allocated, a VLS is fixed sized.}
1347on the allocating stack, provided the allocating stack is large enough.
1348For a VLS stack allocation/deallocation is an inexpensive adjustment of the stack pointer, modulo any stack constructor costs (\eg initial frame setup).
1349For heap stack allocation, allocation/deallocation is an expensive heap allocation (where the heap can be a shared resource), modulo any stack constructor costs.
1350With 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.
1351Currently, \CFA supports stack/heap allocated descriptors but only fixed-sized heap allocated stacks.
1352In \CFA debug-mode, the fixed-sized stack is terminated with a write-only page, which catches most stack overflows.
1353Experience teaching concurrency with \uC~\cite{CS343} shows fixed-sized stacks are rarely an issue for students.
1354Split-stack allocation is under development but requires recompilation of legacy code, which may be impossible.
1355
1356\begin{figure}
1357\centering
1358\input{corlayout.pstex_t}
1359\caption{Coroutine memory layout}
1360\label{f:CoroutineMemoryLayout}
1361\end{figure}
1362
1363
1364\section{Concurrency}
1365\label{s:Concurrency}
1366
1367Concurrency is nondeterministic scheduling of independent sequential execution paths (threads), where each thread has its own stack.
1368A single thread with multiple call stacks, \newterm{coroutining}~\cite{Conway63,Marlin80}, does \emph{not} imply concurrency~\cite[\S~2]{Buhr05a}.
1369In coroutining, coroutines self-schedule the thread across stacks so execution is deterministic.
1370(It is \emph{impossible} to generate a concurrency error when coroutining.)
1371However, coroutines are a stepping stone towards concurrency.
1372
1373The 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}.
1374Therefore, a minimal concurrency system requires coroutines \emph{in conjunction with a nondeterministic scheduler}.
1375The resulting execution system now follows a cooperative threading model~\cite{Adya02,libdill}, called \newterm{non-preemptive scheduling}.
1376Adding \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}.
1377While a scheduler introduces uncertain execution among explicit context switches, preemption introduces uncertainty by introducing implicit context switches.
1378Uncertainty gives the illusion of parallelism on a single processor and provides a mechanism to access and increase performance on multiple processors.
1379The reason is that the scheduler/runtime have complete knowledge about resources and how to best utilized them.
1380However, the introduction of unrestricted nondeterminism results in the need for \newterm{mutual exclusion} and \newterm{synchronization}, which restrict nondeterminism for correctness;
1381otherwise, it is impossible to write meaningful concurrent programs.
1382Optimal concurrent performance is often obtained by having as much nondeterminism as mutual exclusion and synchronization correctness allow.
1383
1384A scheduler can either be a stackless or stackful.
1385For 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.
1386For 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.
1387The \CFA runtime uses a stackful scheduler for uniformity and security.
1388
1389
1390\subsection{Thread}
1391\label{s:threads}
1392
1393Threading needs the ability to start a thread and wait for its completion.
1394A common API for this ability is @fork@ and @join@.
1395\begin{cquote}
1396\begin{tabular}{@{}lll@{}}
1397\multicolumn{1}{c}{\textbf{Java}} & \multicolumn{1}{c}{\textbf{\Celeven}} & \multicolumn{1}{c}{\textbf{pthreads}} \\
1398\begin{cfa}
1399class MyTask extends Thread {...}
1400mytask t = new MyTask(...);
1401`t.start();` // start
1402// concurrency
1403`t.join();` // wait
1404\end{cfa}
1405&
1406\begin{cfa}
1407class MyTask { ... } // functor
1408MyTask mytask;
1409`thread t( mytask, ... );` // start
1410// concurrency
1411`t.join();` // wait
1412\end{cfa}
1413&
1414\begin{cfa}
1415void * rtn( void * arg ) {...}
1416pthread_t t;  int i = 3;
1417`pthread_create( &t, rtn, (void *)i );` // start
1418// concurrency
1419`pthread_join( t, NULL );` // wait
1420\end{cfa}
1421\end{tabular}
1422\end{cquote}
1423\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.
1424\begin{cfa}
1425thread MyTask {};
1426void main( MyTask & this ) { ... }
1427int main() {
1428        MyTask team`[10]`; $\C[2.5in]{// allocate stack-based threads, implicit start after construction}$
1429        // concurrency
1430} $\C{// deallocate stack-based threads, implicit joins before destruction}$
1431\end{cfa}
1432This semantic ensures a thread is started and stopped exactly once, eliminating some programming error, and scales to multiple threads for basic (termination) synchronization.
1433For block allocation to arbitrary depth, including recursion, threads are created/destroyed in a lattice structure (tree with top and bottom).
1434Arbitrary topologies are possible using dynamic allocation, allowing threads to outlive their declaration scope, identical to normal dynamic allocation.
1435\begin{cfa}
1436MyTask * factory( int N ) { ... return `anew( N )`; } $\C{// allocate heap-based threads, implicit start after construction}$
1437int main() {
1438        MyTask * team = factory( 10 );
1439        // concurrency
1440        `delete( team );` $\C{// deallocate heap-based threads, implicit joins before destruction}\CRT$
1441}
1442\end{cfa}
1443
1444Figure~\ref{s:ConcurrentMatrixSummation} shows concurrently adding the rows of a matrix and then totalling the subtotals sequentially, after all the row threads have terminated.
1445The program uses heap-based threads because each thread needs different constructor values.
1446(Python provides a simple iteration mechanism to initialize array elements to different values allowing stack allocation.)
1447The allocation/deallocation pattern appears unusual because allocated objects are immediately deallocated without any intervening code.
1448However, for threads, the deletion provides implicit synchronization, which is the intervening code.
1449% 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.
1450
1451\begin{figure}
1452\begin{cfa}
1453`thread` Adder { int * row, cols, & subtotal; } $\C{// communication variables}$
1454void ?{}( Adder & adder, int row[], int cols, int & subtotal ) {
1455        adder.[ row, cols, &subtotal ] = [ row, cols, &subtotal ];
1456}
1457void main( Adder & adder ) with( adder ) {
1458        subtotal = 0;
1459        for ( c; cols ) { subtotal += row[c]; }
1460}
1461int main() {
1462        const int rows = 10, cols = 1000;
1463        int matrix[rows][cols], subtotals[rows], total = 0;
1464        // read matrix
1465        Adder * adders[rows];
1466        for ( r; rows; ) { $\C{// start threads to sum rows}$
1467                adders[r] = `new( matrix[r], cols, &subtotals[r] );`
1468        }
1469        for ( r; rows ) { $\C{// wait for threads to finish}$
1470                `delete( adders[r] );` $\C{// termination join}$
1471                total += subtotals[r]; $\C{// total subtotal}$
1472        }
1473        sout | total;
1474}
1475\end{cfa}
1476\caption{Concurrent matrix summation}
1477\label{s:ConcurrentMatrixSummation}
1478\end{figure}
1479
1480
1481\subsection{Thread Implementation}
1482
1483Threads 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.
1484Like coroutines, and for the same design reasons, \CFA provides a custom @thread@ type and a @trait@ to enforce and restrict the task-interface functions.
1485\begin{cquote}
1486\begin{tabular}{@{}c@{\hspace{3\parindentlnth}}c@{}}
1487\begin{cfa}
1488thread myThread {
1489        ... // declaration/communication variables
1490};
1491
1492
1493\end{cfa}
1494&
1495\begin{cfa}
1496trait is_thread( `dtype` T ) {
1497        void main( T & );
1498        thread_desc * get_thread( T & );
1499        void ^?{}( T & `mutex` );
1500};
1501\end{cfa}
1502\end{tabular}
1503\end{cquote}
1504Like 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).
1505Similarly, 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.
1506(The qualifier @mutex@ for the destructor parameter is discussed in Section~\ref{s:Monitor}.)
1507The 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;
1508whereas, a thread is scheduling for execution in @main@ immediately after its constructor is run.
1509No 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.
1510
1511
1512\section{Mutual Exclusion / Synchronization}
1513
1514Unrestricted nondeterminism is meaningless as there is no way to know when the result is completed without synchronization.
1515To 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}.
1516Some 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).
1517However, these approaches introduce a new communication mechanism for concurrency different from the standard communication using function call/return.
1518Hence, a programmer must learn and manipulate two sets of design/programming patterns.
1519While this distinction can be hidden away in library code, effective use of the library still has to take both paradigms into account.
1520In contrast, approaches based on stateful models more closely resemble the standard call/return programming model, resulting in a single programming paradigm.
1521
1522At 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}.
1523However, for productivity it is always desirable to use the highest-level construct that provides the necessary efficiency~\cite{Hochstein05}.
1524A newer approach for restricting non-determinism is transactional memory~\cite{Herlihy93}.
1525While 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.
1526
1527One of the most natural, elegant, and efficient mechanisms for mutual exclusion and synchronization for shared-memory systems is the \emph{monitor}.
1528First 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}.
1529In 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.
1530For these reasons, \CFA selected monitors as the core high-level concurrency construct, upon which higher-level approaches can be easily constructed.
1531
1532
1533\subsection{Mutual Exclusion}
1534
1535A 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}.
1536The 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.
1537The readers/writer problem~\cite{Courtois71} is an instance of a group critical-section, where readers share a session but writers have a unique session.
1538
1539However, many solutions exist for mutual exclusion, which vary in terms of performance, flexibility and ease of use.
1540Methods 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.
1541Ease 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.
1542For 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.
1543However, a significant challenge with locks is composability because it takes careful organization for multiple locks to be used while preventing deadlock.
1544Easing composability is another feature higher-level mutual-exclusion mechanisms can offer.
1545
1546
1547\subsection{Synchronization}
1548
1549Synchronization enforces relative ordering of execution, and synchronization tools provide numerous mechanisms to establish these timing relationships.
1550Low-level synchronization primitives offer good performance and flexibility at the cost of ease of use;
1551higher-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.
1552Often 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.
1553If the calling reader is scheduled before the waiting writer, the reader has \newterm{barged}.
1554Barging 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).
1555Preventing or detecting barging is an involved challenge with low-level locks, which is made easier through higher-level constructs.
1556This challenge is often split into two different approaches: barging avoidance and prevention.
1557Algorithms that unconditionally releasing a lock for competing threads to acquire use barging avoidance during synchronization to force a barging thread to wait;
1558algorithms that conditionally hold locks during synchronization, \eg baton-passing~\cite{Andrews89}, prevent barging completely.
1559
1560
1561\section{Monitor}
1562\label{s:Monitor}
1563
1564A \textbf{monitor} is a set of functions that ensure mutual exclusion when accessing shared state.
1565More 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).
1566Restricting acquire/release points eases programming, comprehension, and maintenance, at a slight cost in flexibility and efficiency.
1567\CFA uses a custom @monitor@ type and leverages declaration semantics (deallocation) to protect active or waiting threads in a monitor.
1568
1569The following is a \CFA monitor implementation of an atomic counter.
1570\begin{cfa}[morekeywords=nomutex]
1571`monitor` Aint { int cnt; }; $\C[4.25in]{// atomic integer counter}$
1572int ++?( Aint & `mutex`$\(_{opt}\)$ this ) with( this ) { return ++cnt; } $\C{// increment}$
1573int ?=?( Aint & `mutex`$\(_{opt}\)$ lhs, int rhs ) with( lhs ) { cnt = rhs; } $\C{// conversions with int}\CRT$
1574int ?=?( int & lhs, Aint & `mutex`$\(_{opt}\)$ rhs ) with( rhs ) { lhs = cnt; }
1575\end{cfa}
1576% 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.
1577% (While a constructor may publish its address into a global variable, doing so generates a race-condition.)
1578The 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.
1579The assignment operators provide bidirectional conversion between an atomic and normal integer without accessing field @cnt@;
1580these operations only need @mutex@, if reading/writing the implementation type is not atomic.
1581The atomic counter is used without any explicit mutual-exclusion and provides thread-safe semantics, which is similar to the \CC template @std::atomic@.
1582\begin{cfa}
1583int i = 0, j = 0, k = 5;
1584Aint x = { 0 }, y = { 0 }, z = { 5 }; $\C{// no mutex required}$
1585++x; ++y; ++z; $\C{// safe increment by multiple threads}$
1586x = 2; y = i; z = k; $\C{// conversions}$
1587i = x; j = y; k = z;
1588\end{cfa}
1589
1590\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.
1591\begin{cfa}
1592monitor M { ... } m;
1593void foo( M & mutex m ) { ... } $\C{// acquire mutual exclusion}$
1594void bar( M & mutex m ) { $\C{// acquire mutual exclusion}$
1595        ... `bar( m );` ... `foo( m );` ... $\C{// reacquire mutual exclusion}$
1596}
1597\end{cfa}
1598\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.
1599Similar safety is offered by \emph{explicit} mechanisms like \CC RAII;
1600monitor \emph{implicit} safety ensures no programmer usage errors.
1601Furthermore, 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;
1602RAII is purely a mutual-exclusion mechanism (see Section~\ref{s:Scheduling}).
1603
1604
1605\subsection{Monitor Implementation}
1606
1607For the same design reasons, \CFA provides a custom @monitor@ type and a @trait@ to enforce and restrict the monitor-interface functions.
1608\begin{cquote}
1609\begin{tabular}{@{}c@{\hspace{3\parindentlnth}}c@{}}
1610\begin{cfa}
1611monitor M {
1612        ... // shared data
1613};
1614
1615\end{cfa}
1616&
1617\begin{cfa}
1618trait is_monitor( `dtype` T ) {
1619        monitor_desc * get_monitor( T & );
1620        void ^?{}( T & mutex );
1621};
1622\end{cfa}
1623\end{tabular}
1624\end{cquote}
1625The @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).
1626% 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.
1627% 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.
1628Similarly, 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.
1629The custom monitor type also inserts any locks needed to implement the mutual exclusion semantics.
1630
1631
1632\subsection{Mutex Acquisition}
1633\label{s:MutexAcquisition}
1634
1635While the monitor lock provides mutual exclusion for shared data, there are implementation options for when and where the locking/unlocking occurs.
1636(Much of this discussion also applies to basic locks.)
1637For example, a monitor may be passed through multiple helper functions before it is necessary to acquire the monitor's mutual exclusion.
1638
1639The benefit of mandatory monitor qualifiers is self-documentation, but requiring both @mutex@ and \lstinline[morekeywords=nomutex]@nomutex@ for all monitor parameters is redundant.
1640Instead, the semantics has one qualifier as the default and the other required.
1641For example, make the safe @mutex@ qualifier the default because assuming \lstinline[morekeywords=nomutex]@nomutex@ may cause subtle errors.
1642Alternatively, make the unsafe \lstinline[morekeywords=nomutex]@nomutex@ qualifier the default because it is the \emph{normal} parameter semantics while @mutex@ parameters are rare.
1643Providing a default qualifier implies knowing whether a parameter is a monitor.
1644Since \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.
1645For 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@.
1646
1647The next semantic decision is establishing which parameter \emph{types} may be qualified with @mutex@.
1648The following has monitor parameter types that are composed of multiple objects.
1649\begin{cfa}
1650monitor M { ... }
1651int f1( M & mutex m ); $\C{// single parameter object}$
1652int f2( M * mutex m ); $\C{// single or multiple parameter object}$
1653int f3( M * mutex m[$\,$] ); $\C{// multiple parameter object}$
1654int f4( stack( M * ) & mutex m ); $\C{// multiple parameters object}$
1655\end{cfa}
1656Function @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.
1657Function @f3@ has a multiple object matrix, and @f4@ a multiple object data structure.
1658While shown shortly, multiple object acquisition is possible, but the number of objects must be statically known.
1659Therefore, \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.
1660
1661For object-oriented monitors, \eg Java, calling a mutex member \emph{implicitly} acquires mutual exclusion of the receiver object, @`rec`.foo(...)@.
1662\CFA has no receiver, and hence, the explicit @mutex@ qualifier is used to specify which objects acquire mutual exclusion.
1663A positive consequence of this design decision is the ability to support multi-monitor functions,\footnote{
1664While object-oriented monitors can be extended with a mutex qualifier for multiple-monitor members, no prior example of this feature could be found.}
1665called \newterm{bulk acquire}.
1666\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.
1667Figure~\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.
1668A \CFA programmer only has to manage when to acquire mutual exclusion;
1669a \CC programmer must select the correct lock and acquisition mechanism from a panoply of locking options.
1670Making good choices for common cases in \CFA simplifies the programming experience and enhances safety.
1671
1672\begin{figure}
1673\centering
1674\begin{lrbox}{\myboxA}
1675\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1676monitor BankAccount {
1677
1678        int balance;
1679} b1 = { 0 }, b2 = { 0 };
1680void deposit( BankAccount & `mutex` b,
1681                        int deposit ) with(b) {
1682        balance += deposit;
1683}
1684void transfer( BankAccount & `mutex` my,
1685        BankAccount & `mutex` your, int me2you ) {
1686
1687        deposit( my, -me2you ); // debit
1688        deposit( your, me2you ); // credit
1689}
1690`thread` Person { BankAccount & b1, & b2; };
1691void main( Person & person ) with(person) {
1692        for ( 10_000_000 ) {
1693                if ( random() % 3 ) deposit( b1, 3 );
1694                if ( random() % 3 ) transfer( b1, b2, 7 );
1695        }
1696}   
1697int main() {
1698        `Person p1 = { b1, b2 }, p2 = { b2, b1 };`
1699
1700} // wait for threads to complete
1701\end{cfa}
1702\end{lrbox}
1703
1704\begin{lrbox}{\myboxB}
1705\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1706struct BankAccount {
1707        `recursive_mutex m;`
1708        int balance = 0;
1709} b1, b2;
1710void deposit( BankAccount & b, int deposit ) {
1711        `scoped_lock lock( b.m );`
1712        b.balance += deposit;
1713}
1714void transfer( BankAccount & my,
1715                        BankAccount & your, int me2you ) {
1716        `scoped_lock lock( my.m, your.m );`
1717        deposit( my, -me2you ); // debit
1718        deposit( your, me2you ); // credit
1719}
1720
1721void person( BankAccount & b1, BankAccount & b2 ) {
1722        for ( int i = 0; i < 10$'$000$'$000; i += 1 ) {
1723                if ( random() % 3 ) deposit( b1, 3 );
1724                if ( random() % 3 ) transfer( b1, b2, 7 );
1725        }
1726}   
1727int main() {
1728        `thread p1(person, ref(b1), ref(b2)), p2(person, ref(b2), ref(b1));`
1729        `p1.join(); p2.join();`
1730}
1731\end{cfa}
1732\end{lrbox}
1733
1734\subfloat[\CFA]{\label{f:CFABank}\usebox\myboxA}
1735\hspace{3pt}
1736\vrule
1737\hspace{3pt}
1738\subfloat[\CC]{\label{f:C++Bank}\usebox\myboxB}
1739\hspace{3pt}
1740\caption{Bank transfer problem}
1741\label{f:BankTransfer}
1742\end{figure}
1743
1744Users can still force the acquiring order by using @mutex@/\lstinline[morekeywords=nomutex]@nomutex@.
1745\begin{cfa}
1746void foo( M & mutex m1, M & mutex m2 ); $\C{// acquire m1 and m2}$
1747void bar( M & mutex m1, M & /* nomutex */ m2 ) { $\C{// acquire m1}$
1748        ... foo( m1, m2 ); ... $\C{// acquire m2}$
1749}
1750void baz( M & /* nomutex */ m1, M & mutex m2 ) { $\C{// acquire m2}$
1751        ... foo( m1, m2 ); ... $\C{// acquire m1}$
1752}
1753\end{cfa}
1754The bulk-acquire semantics allow @bar@ or @baz@ to acquire a monitor lock and reacquire it in @foo@.
1755The calls to @bar@ and @baz@ acquired the monitors in opposite order, possibly resulting in deadlock.
1756However, this case is the simplest instance of the \emph{nested-monitor problem}~\cite{Lister77}, where monitors are acquired in sequence versus bulk.
1757Detecting the nested-monitor problem requires dynamic tracking of monitor calls, and dealing with it requires rollback semantics~\cite{Dice10}.
1758\CFA does not deal with this fundamental problem.
1759
1760Finally, like Java, \CFA offers an alternative @mutex@ statement to reduce refactoring and naming.
1761\begin{cquote}
1762\renewcommand{\arraystretch}{0.0}
1763\begin{tabular}{@{}l@{\hspace{3\parindentlnth}}l@{}}
1764\multicolumn{1}{c}{\textbf{\lstinline@mutex@ call}} & \multicolumn{1}{c}{\lstinline@mutex@ \textbf{statement}} \\
1765\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1766monitor M { ... };
1767void foo( M & mutex m1, M & mutex m2 ) {
1768        // critical section
1769}
1770void bar( M & m1, M & m2 ) {
1771        foo( m1, m2 );
1772}
1773\end{cfa}
1774&
1775\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1776
1777void bar( M & m1, M & m2 ) {
1778        mutex( m1, m2 ) {       // remove refactoring and naming
1779                // critical section
1780        }
1781}
1782
1783\end{cfa}
1784\end{tabular}
1785\end{cquote}
1786
1787
1788\subsection{Scheduling}
1789\label{s:Scheduling}
1790
1791% 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.
1792% 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.
1793This 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.)
1794While 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.
1795Leaving the monitor and trying again (busy waiting) is impractical for high-level programming.
1796Monitors eliminate busy waiting by providing synchronization to schedule threads needing access to the shared data, where threads block versus spinning.
1797Synchronization is generally achieved with internal~\cite{Hoare74} or external~\cite[\S~2.9.2]{uC++} scheduling.
1798\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.
1799Finally, \CFA monitors do not allow calling threads to barge ahead of signalled threads, which simplifies synchronization among threads in the monitor and increases correctness.
1800If barging is allowed, synchronization between a signaller and signallee is difficult, often requiring additional flags and multiple unblock/block cycles.
1801In fact, signals-as-hints is completely opposite from that proposed by Hoare in the seminal paper on monitors~\cite[p.~550]{Hoare74}.
1802% \begin{cquote}
1803% 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.
1804% 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}
1805% \end{cquote}
1806Furthermore, \CFA concurrency has no spurious wakeup~\cite[\S~9]{Buhr05a}, which eliminates an implicit form of self barging.
1807Hence, a \CFA @wait@ statement is not enclosed in a @while@ loop retesting a blocking predicate, which can cause thread starvation due to barging.
1808
1809Figure~\ref{f:MonitorScheduling} shows general internal/external scheduling (for the bounded-buffer example in Figure~\ref{f:InternalExternalScheduling}).
1810External calling threads block on the calling queue, if the monitor is occupied, otherwise they enter in FIFO order.
1811Internal threads block on condition queues via @wait@ and reenter from the condition in FIFO order.
1812Alternatively, 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.
1813
1814There are three signalling mechanisms to unblock waiting threads to enter the monitor.
1815Note, 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.
1816For internal scheduling, threads are unblocked from condition queues using @signal@, where the signallee is moved to urgent and the signaller continues (solid line).
1817Multiple signals move multiple signallees to urgent until the condition is empty.
1818When the signaller exits or waits, a thread blocked on urgent is processed before calling threads to prevent barging.
1819(Java conceptually moves the signalled thread to the calling queue, and hence, allows barging.)
1820The 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.
1821
1822For external scheduling, the condition queues are not used;
1823instead threads are unblocked directly from the calling queue using @waitfor@ based on function names requesting mutual exclusion.
1824(The linear search through the calling queue to locate a particular call can be reduced to $O(1)$.)
1825The @waitfor@ has the same semantics as @signal_block@, where the signalled thread executes before the signallee, which waits on urgent.
1826Executing multiple @waitfor@s from different signalled functions causes the calling threads to move to urgent.
1827External scheduling requires urgent to be a stack, because the signaller expects to execute immediately after the specified monitor call has exited or waited.
1828Internal 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.
1829If 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.
1830We tried both a stack for @waitfor@ and queue for signalling, but that resulted in complex semantics about which thread enters next.
1831Hence, \CFA uses a single urgent stack to correctly handle @waitfor@ and adequately support both forms of signalling.
1832
1833\begin{figure}
1834\centering
1835% \subfloat[Scheduling Statements] {
1836% \label{fig:SchedulingStatements}
1837% {\resizebox{0.45\textwidth}{!}{\input{CondSigWait.pstex_t}}}
1838\input{CondSigWait.pstex_t}
1839% }% subfloat
1840% \quad
1841% \subfloat[Bulk acquire monitor] {
1842% \label{fig:BulkMonitor}
1843% {\resizebox{0.45\textwidth}{!}{\input{ext_monitor.pstex_t}}}
1844% }% subfloat
1845\caption{Monitor Scheduling}
1846\label{f:MonitorScheduling}
1847\end{figure}
1848
1849Figure~\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@.
1850The @wait@ function atomically blocks the calling thread and implicitly releases the monitor lock(s) for all monitors in the function's parameter list.
1851The appropriate condition variable is signalled to unblock an opposite kind of thread after an element is inserted/removed from the buffer.
1852Signalling is unconditional, because signalling an empty condition variable does nothing.
1853It 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.
1854In \CFA, a condition variable can be created/stored independently.
1855% 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.
1856
1857% Signalling semantics cannot have the signaller and signalled thread in the monitor simultaneously, which means:
1858% \begin{enumerate}
1859% \item
1860% The signalling thread returns immediately and the signalled thread continues.
1861% \item
1862% The signalling thread continues and the signalled thread is marked for urgent unblocking at the next scheduling point (exit/wait).
1863% \item
1864% The signalling thread blocks but is marked for urgent unblocking at the next scheduling point and the signalled thread continues.
1865% \end{enumerate}
1866% The first approach is too restrictive, as it precludes solving a reasonable class of problems, \eg dating service (see Figure~\ref{f:DatingService}).
1867% \CFA supports the next two semantics as both are useful.
1868
1869\begin{figure}
1870\centering
1871\newbox\myboxA
1872\begin{lrbox}{\myboxA}
1873\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1874forall( otype T ) { // distribute forall
1875        monitor Buffer {
1876                `condition` full, empty;
1877                int front, back, count;
1878                T elements[10];
1879        };
1880        void ?{}( Buffer(T) & buffer ) with(buffer) {
1881                front = back = count = 0;
1882        }
1883        void insert( Buffer(T) & mutex buffer, T elem )
1884                                with(buffer) {
1885                if ( count == 10 ) `wait( empty )`;
1886                // insert elem into buffer
1887                `signal( full )`;
1888        }
1889        T remove( Buffer(T) & mutex buffer ) with(buffer) {
1890                if ( count == 0 ) `wait( full )`;
1891                // remove elem from buffer
1892                `signal( empty )`;
1893                return elem;
1894        }
1895}
1896\end{cfa}
1897\end{lrbox}
1898
1899% \newbox\myboxB
1900% \begin{lrbox}{\myboxB}
1901% \begin{cfa}[aboveskip=0pt,belowskip=0pt]
1902% forall( otype T ) { // distribute forall
1903%       monitor Buffer {
1904%
1905%               int front, back, count;
1906%               T elements[10];
1907%       };
1908%       void ?{}( Buffer(T) & buffer ) with(buffer) {
1909%               [front, back, count] = 0;
1910%       }
1911%       T remove( Buffer(T) & mutex buffer ); // forward
1912%       void insert( Buffer(T) & mutex buffer, T elem )
1913%                               with(buffer) {
1914%               if ( count == 10 ) `waitfor( remove, buffer )`;
1915%               // insert elem into buffer
1916%
1917%       }
1918%       T remove( Buffer(T) & mutex buffer ) with(buffer) {
1919%               if ( count == 0 ) `waitfor( insert, buffer )`;
1920%               // remove elem from buffer
1921%
1922%               return elem;
1923%       }
1924% }
1925% \end{cfa}
1926% \end{lrbox}
1927
1928\newbox\myboxB
1929\begin{lrbox}{\myboxB}
1930\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1931monitor ReadersWriter {
1932        int rcnt, wcnt; // readers/writer using resource
1933};
1934void ?{}( ReadersWriter & rw ) with(rw) {
1935        rcnt = wcnt = 0;
1936}
1937void EndRead( ReadersWriter & mutex rw ) with(rw) {
1938        rcnt -= 1;
1939}
1940void EndWrite( ReadersWriter & mutex rw ) with(rw) {
1941        wcnt = 0;
1942}
1943void StartRead( ReadersWriter & mutex rw ) with(rw) {
1944        if ( wcnt > 0 ) `waitfor( EndWrite, rw );`
1945        rcnt += 1;
1946}
1947void StartWrite( ReadersWriter & mutex rw ) with(rw) {
1948        if ( wcnt > 0 ) `waitfor( EndWrite, rw );`
1949        else while ( rcnt > 0 ) `waitfor( EndRead, rw );`
1950        wcnt = 1;
1951}
1952
1953\end{cfa}
1954\end{lrbox}
1955
1956\subfloat[Generic bounded buffer, internal scheduling]{\label{f:BBInt}\usebox\myboxA}
1957\hspace{3pt}
1958\vrule
1959\hspace{3pt}
1960\subfloat[Readers / writer lock, external scheduling]{\label{f:RWExt}\usebox\myboxB}
1961
1962\caption{Internal / external scheduling}
1963\label{f:InternalExternalScheduling}
1964\end{figure}
1965
1966Figure~\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.
1967\begin{cfa}[aboveskip=2pt,belowskip=1pt]
1968if ( count == 10 ) `waitfor( remove, buffer )`;       |      if ( count == 0 ) `waitfor( insert, buffer )`;
1969\end{cfa}
1970Here, 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.
1971External 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.
1972If 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.
1973Threads 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.
1974Figure~\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@.
1975The writer does a similar action for each reader or writer using the resource.
1976Note, no new calls to @StarRead@/@StartWrite@ may occur when waiting for the call to @EndRead@/@EndWrite@.
1977External scheduling allows waiting for events from other threads while restricting unrelated events, that would otherwise have to wait on conditions in the monitor.
1978The 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.
1979While both mechanisms have strengths and weaknesses, this project uses the control-flow mechanism to be consistent with other language features.
1980% 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}).
1981
1982Figure~\ref{f:DatingService} shows a dating service demonstrating non-blocking and blocking signalling.
1983The dating service matches girl and boy threads with matching compatibility codes so they can exchange phone numbers.
1984A thread blocks until an appropriate partner arrives.
1985The complexity is exchanging phone numbers in the monitor because of the mutual-exclusion property.
1986For 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.
1987For 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.
1988The 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;
1989as well, an arriving thread may not find a partner and must wait, which requires a condition variable, and condition variables imply internal scheduling.
1990Furthermore, 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.
1991Putting loops around the @wait@s does not correct the problem;
1992the simple solution must be restructured to account for barging.
1993
1994\begin{figure}
1995\centering
1996\newbox\myboxA
1997\begin{lrbox}{\myboxA}
1998\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1999enum { CCodes = 20 };
2000monitor DS {
2001        int GirlPhNo, BoyPhNo;
2002        condition Girls[CCodes], Boys[CCodes];
2003        `condition exchange;`
2004};
2005int girl( DS & mutex ds, int phNo, int ccode ) {
2006        if ( is_empty( Boys[ccode] ) ) {
2007                wait( Girls[ccode] );
2008                GirlPhNo = phNo;
2009                `signal( exchange );`
2010        } else {
2011                GirlPhNo = phNo;
2012                `signal( Boys[ccode] );`
2013                `wait( exchange );`
2014        }
2015        return BoyPhNo;
2016}
2017int boy( DS & mutex ds, int phNo, int ccode ) {
2018        // as above with boy/girl interchanged
2019}
2020\end{cfa}
2021\end{lrbox}
2022
2023\newbox\myboxB
2024\begin{lrbox}{\myboxB}
2025\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2026
2027monitor DS {
2028        int GirlPhNo, BoyPhNo;
2029        condition Girls[CCodes], Boys[CCodes];
2030
2031};
2032int girl( DS & mutex ds, int phNo, int ccode ) {
2033        if ( is_empty( Boys[ccode] ) ) { // no compatible
2034                wait( Girls[ccode] ); // wait for boy
2035                GirlPhNo = phNo; // make phone number available
2036
2037        } else {
2038                GirlPhNo = phNo; // make phone number available
2039                `signal_block( Boys[ccode] );` // restart boy
2040
2041        } // if
2042        return BoyPhNo;
2043}
2044int boy( DS & mutex ds, int phNo, int ccode ) {
2045        // as above with boy/girl interchanged
2046}
2047\end{cfa}
2048\end{lrbox}
2049
2050\subfloat[\lstinline@signal@]{\label{f:DatingSignal}\usebox\myboxA}
2051\qquad
2052\subfloat[\lstinline@signal_block@]{\label{f:DatingSignalBlock}\usebox\myboxB}
2053\caption{Dating service}
2054\label{f:DatingService}
2055\end{figure}
2056
2057In 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;
2058the 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.
2059The waiter unblocks next from the urgent queue, uses/takes the state, and exits the monitor.
2060Blocking signal is the reverse, where the waiter is providing the cooperation for the signalling thread;
2061the 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.
2062The waiter changes state and exits the monitor, and the signaller unblocks next from the urgent queue to use/take the state.
2063
2064Both internal and external scheduling extend to multiple monitors in a natural way.
2065\begin{cquote}
2066\begin{tabular}{@{}l@{\hspace{3\parindentlnth}}l@{}}
2067\begin{cfa}
2068monitor M { `condition e`; ... };
2069void foo( M & mutex m1, M & mutex m2 ) {
2070        ... wait( `e` ); ...   // wait( e, m1, m2 )
2071        ... wait( `e, m1` ); ...
2072        ... wait( `e, m2` ); ...
2073}
2074\end{cfa}
2075&
2076\begin{cfa}
2077void rtn$\(_1\)$( M & mutex m1, M & mutex m2 );
2078void rtn$\(_2\)$( M & mutex m1 );
2079void bar( M & mutex m1, M & mutex m2 ) {
2080        ... waitfor( `rtn` ); ...       // $\LstCommentStyle{waitfor( rtn\(_1\), m1, m2 )}$
2081        ... waitfor( `rtn, m1` ); ... // $\LstCommentStyle{waitfor( rtn\(_2\), m1 )}$
2082}
2083\end{cfa}
2084\end{tabular}
2085\end{cquote}
2086For @wait( e )@, the default semantics is to atomically block the signaller and release all acquired mutex parameters, \ie @wait( e, m1, m2 )@.
2087To override the implicit multi-monitor wait, specific mutex parameter(s) can be specified, \eg @wait( e, m1 )@.
2088Wait cannot statically verifies the released monitors are the acquired mutex-parameters without disallowing separately compiled helper functions calling @wait@.
2089While \CC supports bulk locking, @wait@ only accepts a single lock for a condition variable, so bulk locking with condition variables is asymmetric.
2090Finally, a signaller,
2091\begin{cfa}
2092void baz( M & mutex m1, M & mutex m2 ) {
2093        ... signal( e ); ...
2094}
2095\end{cfa}
2096must 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.
2097
2098Similarly, for @waitfor( rtn )@, the default semantics is to atomically block the acceptor and release all acquired mutex parameters, \ie @waitfor( rtn, m1, m2 )@.
2099To override the implicit multi-monitor wait, specific mutex parameter(s) can be specified, \eg @waitfor( rtn, m1 )@.
2100@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.
2101% When an overloaded function appears in an @waitfor@ statement, calls to any function with that name are accepted.
2102% The rationale is that members with the same name should perform a similar function, and therefore, all should be eligible to accept a call.
2103Overloaded functions can be disambiguated using a cast
2104\begin{cfa}
2105void rtn( M & mutex m );
2106`int` rtn( M & mutex m );
2107waitfor( (`int` (*)( M & mutex ))rtn, m );
2108\end{cfa}
2109
2110The ability to release a subset of acquired monitors can result in a \newterm{nested monitor}~\cite{Lister77} deadlock.
2111\begin{cfa}
2112void foo( M & mutex m1, M & mutex m2 ) {
2113        ... wait( `e, m1` ); ...                                $\C{// release m1, keeping m2 acquired )}$
2114void bar( M & mutex m1, M & mutex m2 ) {        $\C{// must acquire m1 and m2 )}$
2115        ... signal( `e` ); ...
2116\end{cfa}
2117The @wait@ only releases @m1@ so the signalling thread cannot acquire @m1@ and @m2@ to enter @bar@ and @signal@ the condition.
2118While deadlock can occur with multiple/nesting acquisition, this is a consequence of locks, and by extension monitors, not being perfectly composable.
2119
2120
2121
2122\subsection{Extended \protect\lstinline@waitfor@}
2123
2124Figure~\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.
2125For a @waitfor@ clause to be executed, its @when@ must be true and an outstanding call to its corresponding member(s) must exist.
2126The \emph{conditional-expression} of a @when@ may call a function, but the function must not block or context switch.
2127If 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@.
2128If 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.
2129If there is a @timeout@ clause, it provides an upper bound on waiting.
2130If 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.
2131Hence, the terminating @else@ clause allows a conditional attempt to accept a call without blocking.
2132If both @timeout@ and @else@ clause are present, the @else@ must be conditional, or the @timeout@ is never triggered.
2133
2134\begin{figure}
2135\centering
2136\begin{cfa}
2137`when` ( $\emph{conditional-expression}$ )      $\C{// optional guard}$
2138        waitfor( $\emph{mutex-member-name}$ ) $\emph{statement}$ $\C{// action after call}$
2139`or` `when` ( $\emph{conditional-expression}$ ) $\C{// any number of functions}$
2140        waitfor( $\emph{mutex-member-name}$ ) $\emph{statement}$
2141`or`    ...
2142`when` ( $\emph{conditional-expression}$ ) $\C{// optional guard}$
2143        `timeout` $\emph{statement}$ $\C{// optional terminating timeout clause}$
2144`when` ( $\emph{conditional-expression}$ ) $\C{// optional guard}$
2145        `else`  $\emph{statement}$ $\C{// optional terminating clause}$
2146\end{cfa}
2147\caption{Extended \protect\lstinline@waitfor@}
2148\label{f:ExtendedWaitfor}
2149\end{figure}
2150
2151Note, a group of conditional @waitfor@ clauses is \emph{not} the same as a group of @if@ statements, \eg:
2152\begin{cfa}
2153if ( C1 ) waitfor( mem1 );                       when ( C1 ) waitfor( mem1 );
2154else if ( C2 ) waitfor( mem2 );         or when ( C2 ) waitfor( mem2 );
2155\end{cfa}
2156The left example only accepts @mem1@ if @C1@ is true or only @mem2@ if @C2@ is true.
2157The right example accepts either @mem1@ or @mem2@ if @C1@ and @C2@ are true.
2158
2159An 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@.
2160\begin{cfa}
2161void main( Buffer(T) & buffer ) with(buffer) {
2162        for () {
2163                `waitfor( ^?{}, buffer )` break;
2164                or when ( count != 20 ) waitfor( insert, buffer ) { ... }
2165                or when ( count != 0 ) waitfor( remove, buffer ) { ... }
2166        }
2167        // clean up
2168}
2169\end{cfa}
2170When the program main deallocates the buffer, it first calls the buffer's destructor, which is accepted, the destructor runs, and the buffer is deallocated.
2171However, the buffer thread cannot continue after the destructor call because the object is gone;
2172hence, clean up in @main@ cannot occur, which means destructors for local objects are not run.
2173To 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.
2174Then, the destructor caller unblocks from urgent to deallocate the object.
2175Accepting the destructor is the idiomatic way in \CFA to terminate a thread performing direct communication.
2176
2177
2178\subsection{Bulk Barging Prevention}
2179
2180Figure~\ref{f:BulkBargingPrevention} shows \CFA code where bulk acquire adds complexity to the internal-signalling semantics.
2181The complexity begins at the end of the inner @mutex@ statement, where the semantics of internal scheduling need to be extended for multiple monitors.
2182The problem is that bulk acquire is used in the inner @mutex@ statement where one of the monitors is already acquired.
2183When 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.
2184However, both the signalling and waiting threads W1 and W2 need some subset of monitors @m1@ and @m2@.
2185\begin{cquote}
2186condition c: (order 1) W2(@m2@), W1(@m1@,@m2@)\ \ \ or\ \ \ (order 2) W1(@m1@,@m2@), W2(@m2@) \\
2187S: 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 \\
2188\hspace*{2.75in}$\rightarrow$ rel. @m1@ $\rightarrow$ pass @m1,m2@ unblock W1 (order 1)
2189\end{cquote}
2190
2191\begin{figure}
2192\newbox\myboxA
2193\begin{lrbox}{\myboxA}
2194\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2195monitor M m1, m2;
2196condition c;
2197mutex( m1 ) { // $\LstCommentStyle{\color{red}outer}$
2198        ...
2199        mutex( m1, m2 ) { // $\LstCommentStyle{\color{red}inner}$
2200                ... `signal( c )`; ...
2201                // m1, m2 still acquired
2202        } // $\LstCommentStyle{\color{red}release m2}$
2203        // m1 acquired
2204} // release m1
2205\end{cfa}
2206\end{lrbox}
2207
2208\newbox\myboxB
2209\begin{lrbox}{\myboxB}
2210\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2211
2212
2213mutex( m1 ) {
2214        ...
2215        mutex( m1, m2 ) {
2216                ... `wait( c )`; // release m1, m2
2217                // m1, m2 reacquired
2218        } // $\LstCommentStyle{\color{red}release m2}$
2219        // m1 acquired
2220} // release m1
2221\end{cfa}
2222\end{lrbox}
2223
2224\newbox\myboxC
2225\begin{lrbox}{\myboxC}
2226\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2227
2228
2229mutex( m2 ) {
2230        ... `wait( c )`; // release m2
2231        // m2 reacquired
2232} // $\LstCommentStyle{\color{red}release m2}$
2233
2234
2235
2236
2237\end{cfa}
2238\end{lrbox}
2239
2240\begin{cquote}
2241\subfloat[Signalling Thread (S)]{\label{f:SignallingThread}\usebox\myboxA}
2242\hspace{3\parindentlnth}
2243\subfloat[Waiting Thread (W1)]{\label{f:WaitingThread}\usebox\myboxB}
2244\hspace{2\parindentlnth}
2245\subfloat[Waiting Thread (W2)]{\label{f:OtherWaitingThread}\usebox\myboxC}
2246\end{cquote}
2247\caption{Bulk Barging Prevention}
2248\label{f:BulkBargingPrevention}
2249\end{figure}
2250
2251One 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.
2252However, 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.
2253If W1 waited first, the signaller must retain @m1@ amd @m2@ until completion of the outer mutex statement and then pass both to W1.
2254% Furthermore, there is an execution sequence where the signaller always finds waiter W2, and hence, waiter W1 starves.
2255To support this efficient semantics (and prevent barging), the implementation maintains a list of monitors acquired for each blocked thread.
2256When a signaller exits or waits in a monitor function/statement, the front waiter on urgent is unblocked if all its monitors are released.
2257Implementing a fast subset check for the necessary released monitors is important.
2258% 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.
2259
2260
2261\subsection{Loose Object Definitions}
2262\label{s:LooseObjectDefinitions}
2263
2264In an object-oriented programming language, a class includes an exhaustive list of operations.
2265A new class can add members via static inheritance but the subclass still has an exhaustive list of operations.
2266(Dynamic member adding, \eg JavaScript~\cite{JavaScript}, is not considered.)
2267In 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@.
2268
2269However, in \CFA, monitor functions can be statically added/removed in translation units, making a fast subset check difficult.
2270\begin{cfa}
2271        monitor M { ... }; // common type, included in .h file
2272translation unit 1
2273        void `f`( M & mutex m );
2274        void g( M & mutex m ) { waitfor( `f`, m ); }
2275translation unit 2
2276        void `f`( M & mutex m ); $\C{// replacing f and g for type M in this translation unit}$
2277        void `g`( M & mutex m );
2278        void h( M & mutex m ) { waitfor( `f`, m ) or waitfor( `g`, m ); } $\C{// extending type M in this translation unit}$
2279\end{cfa}
2280The @waitfor@ statements in each translation unit cannot form a unique bit-mask because the monitor type does not carry that information.
2281Hence, function pointers are used to identify the functions listed in the @waitfor@ statement, stored in a variable-sized array.
2282Then, the same implementation approach used for the urgent stack is used for the calling queue.
2283Each 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.
2284(A possible way to construct a dense mapping is at link or load-time.)
2285
2286
2287\subsection{Multi-Monitor Scheduling}
2288\label{s:Multi-MonitorScheduling}
2289
2290External scheduling, like internal scheduling, becomes significantly more complex for multi-monitor semantics.
2291Even in the simplest case, new semantics need to be established.
2292\begin{cfa}
2293monitor M { ... };
2294void f( M & mutex m1 );
2295void g( M & mutex m1, M & mutex m2 ) { `waitfor( f );` } $\C{// pass m1 or m2 to f?}$
2296\end{cfa}
2297The solution is for the programmer to disambiguate:
2298\begin{cfa}
2299waitfor( f, `m2` ); $\C{// wait for call to f with argument m2}$
2300\end{cfa}
2301Both 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@.
2302This behaviour can be extended to the multi-monitor @waitfor@ statement.
2303\begin{cfa}
2304monitor M { ... };
2305void f( M & mutex m1, M & mutex m2 );
2306void g( M & mutex m1, M & mutex m2 ) { waitfor( f, `m1, m2` ); $\C{// wait for call to f with arguments m1 and m2}$
2307\end{cfa}
2308Again, the set of monitors passed to the @waitfor@ statement must be entirely contained in the set of monitors already acquired by the accepting function.
2309Also, the order of the monitors in a @waitfor@ statement is unimportant.
2310
2311Figure~\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.
2312For both examples, the set of monitors is disjoint so unblocking is impossible.
2313
2314\begin{figure}
2315\centering
2316\begin{lrbox}{\myboxA}
2317\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2318monitor M1 {} m11, m12;
2319monitor M2 {} m2;
2320condition c;
2321void f( M1 & mutex m1, M2 & mutex m2 ) {
2322        signal( c );
2323}
2324void g( M1 & mutex m1, M2 & mutex m2 ) {
2325        wait( c );
2326}
2327g( `m11`, m2 ); // block on wait
2328f( `m12`, m2 ); // cannot fulfil
2329\end{cfa}
2330\end{lrbox}
2331
2332\begin{lrbox}{\myboxB}
2333\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2334monitor M1 {} m11, m12;
2335monitor M2 {} m2;
2336
2337void f( M1 & mutex m1, M2 & mutex m2 ) {
2338
2339}
2340void g( M1 & mutex m1, M2 & mutex m2 ) {
2341        waitfor( f, m1, m2 );
2342}
2343g( `m11`, m2 ); // block on accept
2344f( `m12`, m2 ); // cannot fulfil
2345\end{cfa}
2346\end{lrbox}
2347\subfloat[Internal scheduling]{\label{f:InternalScheduling}\usebox\myboxA}
2348\hspace{3pt}
2349\vrule
2350\hspace{3pt}
2351\subfloat[External scheduling]{\label{f:ExternalScheduling}\usebox\myboxB}
2352\caption{Unmatched \protect\lstinline@mutex@ sets}
2353\label{f:UnmatchedMutexSets}
2354\end{figure}
2355
2356
2357\subsection{\protect\lstinline@mutex@ Threads}
2358
2359Threads 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.
2360Hence, all monitor features are available when using threads.
2361Figure~\ref{f:DirectCommunication} shows a comparison of direct call communication in \CFA with direct channel communication in Go.
2362(Ada provides a similar mechanism to the \CFA direct communication.)
2363The program main in both programs communicates directly with the other thread versus indirect communication where two threads interact through a passive monitor.
2364Both direct and indirection thread communication are valuable tools in structuring concurrent programs.
2365
2366\begin{figure}
2367\centering
2368\begin{lrbox}{\myboxA}
2369\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2370
2371struct Msg { int i, j; };
2372thread GoRtn { int i;  float f;  Msg m; };
2373void mem1( GoRtn & mutex gortn, int i ) { gortn.i = i; }
2374void mem2( GoRtn & mutex gortn, float f ) { gortn.f = f; }
2375void mem3( GoRtn & mutex gortn, Msg m ) { gortn.m = m; }
2376void ^?{}( GoRtn & mutex ) {}
2377
2378void main( GoRtn & gortn ) with( gortn ) {  // thread starts
2379
2380        for () {
2381
2382                `waitfor( mem1, gortn )` sout | i;  // wait for calls
2383                or `waitfor( mem2, gortn )` sout | f;
2384                or `waitfor( mem3, gortn )` sout | m.i | m.j;
2385                or `waitfor( ^?{}, gortn )` break;
2386
2387        }
2388
2389}
2390int main() {
2391        GoRtn gortn; $\C[2.0in]{// start thread}$
2392        `mem1( gortn, 0 );` $\C{// different calls}\CRT$
2393        `mem2( gortn, 2.5 );`
2394        `mem3( gortn, (Msg){1, 2} );`
2395
2396
2397} // wait for completion
2398\end{cfa}
2399\end{lrbox}
2400
2401\begin{lrbox}{\myboxB}
2402\begin{Go}[aboveskip=0pt,belowskip=0pt]
2403func main() {
2404        type Msg struct{ i, j int }
2405
2406        ch1 := make( chan int )
2407        ch2 := make( chan float32 )
2408        ch3 := make( chan Msg )
2409        hand := make( chan string )
2410        shake := make( chan string )
2411        gortn := func() { $\C[1.5in]{// thread starts}$
2412                var i int;  var f float32;  var m Msg
2413                L: for {
2414                        select { $\C{// wait for messages}$
2415                          case `i = <- ch1`: fmt.Println( i )
2416                          case `f = <- ch2`: fmt.Println( f )
2417                          case `m = <- ch3`: fmt.Println( m )
2418                          case `<- hand`: break L $\C{// sentinel}$
2419                        }
2420                }
2421                `shake <- "SHAKE"` $\C{// completion}$
2422        }
2423
2424        go gortn() $\C{// start thread}$
2425        `ch1 <- 0` $\C{// different messages}$
2426        `ch2 <- 2.5`
2427        `ch3 <- Msg{1, 2}`
2428        `hand <- "HAND"` $\C{// sentinel value}$
2429        `<- shake` $\C{// wait for completion}\CRT$
2430}
2431\end{Go}
2432\end{lrbox}
2433
2434\subfloat[\CFA]{\label{f:CFAwaitfor}\usebox\myboxA}
2435\hspace{3pt}
2436\vrule
2437\hspace{3pt}
2438\subfloat[Go]{\label{f:Gochannel}\usebox\myboxB}
2439\caption{Direct communication}
2440\label{f:DirectCommunication}
2441\end{figure}
2442
2443\begin{comment}
2444The following shows an example of two threads directly calling each other and accepting calls from each other in a cycle.
2445\begin{cfa}
2446\end{cfa}
2447\vspace{-0.8\baselineskip}
2448\begin{cquote}
2449\begin{tabular}{@{}l@{\hspace{3\parindentlnth}}l@{}}
2450\begin{cfa}
2451thread Ping {} pi;
2452void ping( Ping & mutex ) {}
2453void main( Ping & pi ) {
2454        for ( 10 ) {
2455                `waitfor( ping, pi );`
2456                `pong( po );`
2457        }
2458}
2459int main() {}
2460\end{cfa}
2461&
2462\begin{cfa}
2463thread Pong {} po;
2464void pong( Pong & mutex ) {}
2465void main( Pong & po ) {
2466        for ( 10 ) {
2467                `ping( pi );`
2468                `waitfor( pong, po );`
2469        }
2470}
2471
2472\end{cfa}
2473\end{tabular}
2474\end{cquote}
2475% \lstMakeShortInline@%
2476% \caption{Threads ping/pong using external scheduling}
2477% \label{f:pingpong}
2478% \end{figure}
2479Note, the ping/pong threads are globally declared, @pi@/@po@, and hence, start (and possibly complete) before the program main starts.
2480\end{comment}
2481
2482
2483\subsection{Execution Properties}
2484
2485Table~\ref{t:ObjectPropertyComposition} shows how the \CFA high-level constructs cover 3 fundamental execution properties: thread, stateful function, and mutual exclusion.
2486Case 1 is a basic object, with none of the new execution properties.
2487Case 2 allows @mutex@ calls to Case 1 to protect shared data.
2488Case 3 allows stateful functions to suspend/resume but restricts operations because the state is stackless.
2489Case 4 allows @mutex@ calls to Case 3 to protect shared data.
2490Cases 5 and 6 are the same as 3 and 4 without restriction because the state is stackful.
2491Cases 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.
2492Cases 9 and 10 have a stackful thread without and with @mutex@ calls.
2493For situations where threads do not require direct communication, case 9 provides faster creation/destruction by eliminating @mutex@ setup.
2494
2495\begin{table}
2496\caption{Object property composition}
2497\centering
2498\label{t:ObjectPropertyComposition}
2499\renewcommand{\arraystretch}{1.25}
2500%\setlength{\tabcolsep}{5pt}
2501\begin{tabular}{c|c|l|l}
2502\multicolumn{2}{c|}{object properties} & \multicolumn{2}{c}{mutual exclusion} \\
2503\hline
2504thread  & stateful                              & \multicolumn{1}{c|}{No} & \multicolumn{1}{c}{Yes} \\
2505\hline
2506\hline
2507No              & No                                    & \textbf{1}\ \ \ aggregate type                & \textbf{2}\ \ \ @monitor@ aggregate type \\
2508\hline
2509No              & Yes (stackless)               & \textbf{3}\ \ \ @generator@                   & \textbf{4}\ \ \ @monitor@ @generator@ \\
2510\hline
2511No              & Yes (stackful)                & \textbf{5}\ \ \ @coroutine@                   & \textbf{6}\ \ \ @monitor@ @coroutine@ \\
2512\hline
2513Yes             & No / Yes (stackless)  & \textbf{7}\ \ \ {\color{red}rejected} & \textbf{8}\ \ \ {\color{red}rejected} \\
2514\hline
2515Yes             & Yes (stackful)                & \textbf{9}\ \ \ @thread@                              & \textbf{10}\ \ @monitor@ @thread@ \\
2516\end{tabular}
2517\end{table}
2518
2519
2520\subsection{Low-level Locks}
2521
2522For 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.
2523Some of these low-level mechanism are used in the \CFA runtime, but we strongly advocate using high-level mechanisms whenever possible.
2524
2525
2526% \section{Parallelism}
2527% \label{s:Parallelism}
2528%
2529% Historically, computer performance was about processor speeds.
2530% However, with heat dissipation being a direct consequence of speed increase, parallelism is the new source for increased performance~\cite{Sutter05, Sutter05b}.
2531% Therefore, high-performance applications must care about parallelism, which requires concurrency.
2532% The lowest-level approach of parallelism is to use \newterm{kernel threads} in combination with semantics like @fork@, @join@, \etc.
2533% However, kernel threads are better as an implementation tool because of complexity and higher cost.
2534% Therefore, different abstractions are often layered onto kernel threads to simplify them, \eg pthreads.
2535%
2536%
2537% \subsection{User Threads}
2538%
2539% A direct improvement on kernel threads is user threads, \eg Erlang~\cite{Erlang} and \uC~\cite{uC++book}.
2540% 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.
2541% In many cases, user threads can be used on a much larger scale (100,000 threads).
2542% Like kernel threads, user threads support preemption, which maximizes nondeterminism, but increases the potential for concurrency errors: race, livelock, starvation, and deadlock.
2543% \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}.
2544%
2545% A variant of user thread is \newterm{fibres}, which removes preemption, \eg Go~\cite{Go} @goroutine@s.
2546% 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.
2547% However, preemption is necessary for fairness and to reduce tail-latency.
2548% For concurrency that relies on spinning, if all cores spin the system is livelocked, whereas preemption breaks the livelock.
2549
2550
2551\begin{comment}
2552\subsection{Thread Pools}
2553
2554In 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.
2555If the jobs are dependent, \ie interact, there is an implicit/explicit dependency graph that ties them together.
2556While removing direct concurrency, and hence the amount of context switching, thread pools significantly limit the interaction that can occur among jobs.
2557Indeed, jobs should not block because that also blocks the underlying thread, which effectively means the CPU utilization, and therefore throughput, suffers.
2558While 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.
2559As well, concurrency errors return, which threads pools are suppose to mitigate.
2560
2561\begin{figure}
2562\centering
2563\begin{tabular}{@{}l|l@{}}
2564\begin{cfa}
2565struct Adder {
2566    int * row, cols;
2567};
2568int operator()() {
2569        subtotal = 0;
2570        for ( int c = 0; c < cols; c += 1 )
2571                subtotal += row[c];
2572        return subtotal;
2573}
2574void ?{}( Adder * adder, int row[$\,$], int cols, int & subtotal ) {
2575        adder.[rows, cols, subtotal] = [rows, cols, subtotal];
2576}
2577
2578
2579
2580
2581\end{cfa}
2582&
2583\begin{cfa}
2584int main() {
2585        const int rows = 10, cols = 10;
2586        int matrix[rows][cols], subtotals[rows], total = 0;
2587        // read matrix
2588        Executor executor( 4 ); // kernel threads
2589        Adder * adders[rows];
2590        for ( r; rows ) { // send off work for executor
2591                adders[r] = new( matrix[r], cols, &subtotal[r] );
2592                executor.send( *adders[r] );
2593        }
2594        for ( r; rows ) {       // wait for results
2595                delete( adders[r] );
2596                total += subtotals[r];
2597        }
2598        sout | total;
2599}
2600\end{cfa}
2601\end{tabular}
2602\caption{Executor}
2603\end{figure}
2604\end{comment}
2605
2606
2607\section{\protect\CFA Runtime Structure}
2608\label{s:CFARuntimeStructure}
2609
2610Figure~\ref{f:RunTimeStructure} illustrates the runtime structure of a \CFA program.
2611In 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.
2612An executing thread is illustrated by its containment in a processor.
2613
2614\begin{figure}
2615\centering
2616\input{RunTimeStructure}
2617\caption{\CFA Runtime structure}
2618\label{f:RunTimeStructure}
2619\end{figure}
2620
2621
2622\subsection{Cluster}
2623\label{s:RuntimeStructureCluster}
2624
2625A \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).
2626The purpose of a cluster is to control the amount of parallelism that is possible among threads, plus scheduling and other execution defaults.
2627The default cluster-scheduler is single-queue multi-server, which provides automatic load-balancing of threads on processors.
2628However, the design allows changing the scheduler, \eg multi-queue multi-server with work-stealing/sharing across the virtual processors.
2629If several clusters exist, both threads and virtual processors, can be explicitly migrated from one cluster to another.
2630No automatic load balancing among clusters is performed by \CFA.
2631
2632When 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.
2633The user cluster is created to contain the application user-threads.
2634Having all threads execute on the one cluster often maximizes utilization of processors, which minimizes runtime.
2635However, because of limitations of scheduling requirements (real-time), NUMA architecture, heterogeneous hardware, or issues with the underlying operating system, multiple clusters are sometimes necessary.
2636
2637
2638\subsection{Virtual Processor}
2639\label{s:RuntimeStructureProcessor}
2640
2641A 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.
2642Programs may use more virtual processors than hardware processors.
2643On a multiprocessor, kernel threads are distributed across the hardware processors resulting in virtual processors executing in parallel.
2644(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.)
2645The \CFA runtime attempts to block unused processors and unblock processors as the system load increases;
2646balancing the workload with processors is difficult because it requires future knowledge, \ie what will the applicaton workload do next.
2647Preemption occurs on virtual processors rather than user threads, via operating-system interrupts.
2648Thus virtual processors execute user threads, where preemption frequency applies to a virtual processor, so preemption occurs randomly across the executed user threads.
2649Turning off preemption transforms user threads into fibres.
2650
2651
2652\begin{comment}
2653\section{Implementation}
2654\label{s:Implementation}
2655
2656A primary implementation challenge is avoiding contention from dynamically allocating memory because of bulk acquire, \eg the internal-scheduling design is (almost) free of allocations.
2657All 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.
2658Furthermore, several bulk-acquire operations need a variable amount of memory.
2659This 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.
2660
2661In \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.
2662When a mutex call is made, pointers to the concerned monitors are aggregated into a variable-length array and sorted.
2663This array persists for the entire duration of the mutual exclusion and is used extensively for synchronization operations.
2664
2665To 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;
2666the corresponding registers are then restored for the other context.
2667Note, the instruction pointer is untouched since the context switch is always inside the same function.
2668Experimental 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.
2669
2670All kernel threads (@pthreads@) created a stack.
2671Each \CFA virtual processor is implemented as a coroutine and these coroutines run directly on the kernel-thread stack, effectively stealing this stack.
2672The exception to this rule is the program main, \ie the initial kernel thread that is given to any program.
2673In 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.
2674\end{comment}
2675
2676
2677\subsection{Preemption}
2678
2679Nondeterministic 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.
2680This atomic reliance can fail on multi-core machines, because execution across cores is nondeterministic.
2681A 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).
2682Preemption is normally handled by setting a countdown timer on each virtual processor.
2683When 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.
2684Multiple signal handlers may be pending.
2685When 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.
2686The 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;
2687therefore, the same signal mask is required for all virtual processors in a cluster.
2688Because preemption frequency is usually long (1 millisecond) performance cost is negligible.
2689
2690Linux switched a decade ago from specific to arbitrary process signal-delivery for applications with multiple kernel threads.
2691\begin{cquote}
2692A process-directed signal may be delivered to any one of the threads that does not currently have the signal blocked.
2693If more than one of the threads has the signal unblocked, then the kernel chooses an arbitrary thread to which it will deliver the signal.
2694SIGNAL(7) - Linux Programmer's Manual
2695\end{cquote}
2696Hence, 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).
2697To 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.
2698Virtual processors register an expiration time with the discrete-event simulator, which is inserted in sorted order.
2699The 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.
2700Processing a preemption event sends an \emph{internal} @SIGUSR1@ signal to the registered virtual processor, which is always delivered to that processor.
2701
2702
2703\subsection{Debug Kernel}
2704
2705There are two versions of the \CFA runtime kernel: debug and non-debug.
2706The 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.
2707After a program is debugged, the non-debugging version can be used to significantly decrease space and increase performance.
2708
2709
2710\section{Performance}
2711\label{results}
2712
2713To 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.
2714For 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.
2715The 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.
2716
2717All benchmarks are run using the following harness.
2718\begin{cfa}
2719unsigned int N = 10_000_000;
2720#define BENCH( `run` ) Time before = getTimeNsec();  `run;`  Duration result = (getTimeNsec() - before) / N;
2721\end{cfa}
2722The method used to get time is @clock_gettime( CLOCK_REALTIME )@.
2723Each benchmark is performed @N@ times, where @N@ varies depending on the benchmark;
2724the total time is divided by @N@ to obtain the average time for a benchmark.
2725Each benchmark experiment is run 31 times.
2726All omitted tests for other languages are functionally identical to the \CFA tests and available online~\cite{CforallBenchMarks}.
2727% tar --exclude=.deps --exclude=Makefile --exclude=Makefile.in --exclude=c.c --exclude=cxx.cpp --exclude=fetch_add.c -cvhf benchmark.tar benchmark
2728
2729\paragraph{Object Creation}
2730
2731Object creation is measured by creating/deleting the specific kind of concurrent object.
2732Figure~\ref{f:creation} shows the code for \CFA, with results in Table~\ref{tab:creation}.
2733The 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.
2734
2735\begin{multicols}{2}
2736\lstset{language=CFA,moredelim=**[is][\color{red}]{@}{@},deletedelim=**[is][]{`}{`}}
2737\begin{cfa}
2738@thread@ MyThread {};
2739void @main@( MyThread & ) {}
2740int main() {
2741        BENCH( for ( N ) { @MyThread m;@ } )
2742        sout | result`ns;
2743}
2744\end{cfa}
2745\captionof{figure}{\CFA object-creation benchmark}
2746\label{f:creation}
2747
2748\columnbreak
2749
2750\vspace*{-16pt}
2751\captionof{table}{Object creation comparison (nanoseconds)}
2752\label{tab:creation}
2753
2754\begin{tabular}[t]{@{}r*{3}{D{.}{.}{5.2}}@{}}
2755\multicolumn{1}{@{}c}{} & \multicolumn{1}{c}{Median} & \multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
2756\CFA Coroutine Lazy             & 14.3          & 14.3          & 0.32          \\
2757\CFA Coroutine Eager    & 522.8         & 525.3         & 5.81          \\
2758\CFA Thread                             & 1257.8        & 1291.2        & 86.19         \\
2759\uC Coroutine                   & 92.2          & 91.4          & 1.58          \\
2760\uC Thread                              & 499.5         & 500.1         & 5.67          \\
2761Goroutine                               & 4397.0        & 4362.8        & 390.77        \\
2762Java Thread                             & 107405.0      & 107794.8      & 1601.33       \\
2763% Qthreads                              & 159.9         & 159.6         & 0.73          \\
2764Pthreads                                & 32920.9       & 32882.7       & 213.55
2765\end{tabular}
2766\end{multicols}
2767
2768
2769\paragraph{Internal Scheduling}
2770
2771Internal scheduling is measured using a cycle of two threads signalling and waiting.
2772Figure~\ref{f:int-sched} shows the code for \CFA, with results in Table~\ref{tab:int-sched}.
2773Note, the incremental cost of bulk acquire for \CFA, which is largely a fixed cost for small numbers of mutex objects.
2774Java 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.
2775
2776\begin{multicols}{2}
2777\lstset{language=CFA,moredelim=**[is][\color{red}]{@}{@},deletedelim=**[is][]{`}{`}}
2778\begin{cfa}
2779volatile int go = 0;
2780@monitor@ M { @condition c;@ } m;
2781void __attribute__((noinline))
2782do_call( M & @mutex@ a1 ) { @signal( c );@ }
2783thread T {};
2784void main( T & this ) {
2785        while ( go == 0 ) { yield(); }
2786        while ( go == 1 ) { do_call( m ); }
2787}
2788int  __attribute__((noinline))
2789do_wait( M & mutex m ) with(m) {
2790        go = 1; // continue other thread
2791        BENCH( for ( N ) { @wait( c );@ } );
2792        go = 0; // stop other thread
2793        sout | result`ns;
2794}
2795int main() {
2796        T t;
2797        do_wait( m );
2798}
2799\end{cfa}
2800\captionof{figure}{\CFA Internal-scheduling benchmark}
2801\label{f:int-sched}
2802
2803\columnbreak
2804
2805\vspace*{-16pt}
2806\captionof{table}{Internal-scheduling comparison (nanoseconds)}
2807\label{tab:int-sched}
2808\bigskip
2809
2810\begin{tabular}{@{}r*{3}{D{.}{.}{5.2}}@{}}
2811\multicolumn{1}{@{}c}{} & \multicolumn{1}{c}{Median} & \multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
2812\CFA @signal@, 1 @monitor@      & 367.0         & 371.5         & 17.34         \\
2813\CFA @signal@, 2 @monitor@      & 477.2         & 478.6         & 8.31          \\
2814\CFA @signal@, 4 @monitor@      & 725.8         & 734.0         & 17.98         \\
2815\uC @signal@                            & 322.8         & 323.0         & 3.64          \\
2816Java @notify@                           & 16520.0       & 20096.7       & 9378.53       \\
2817Pthreads Cond. Variable         & 4931.3        & 5057.0        & 326.80
2818\end{tabular}
2819\end{multicols}
2820
2821
2822\paragraph{External Scheduling}
2823
2824External scheduling is measured using a cycle of two threads calling and accepting the call using the @waitfor@ statement.
2825Figure~\ref{f:ext-sched} shows the code for \CFA, with results in Table~\ref{tab:ext-sched}.
2826Note, the incremental cost of bulk acquire for \CFA, which is largely a fixed cost for small numbers of mutex objects.
2827
2828\begin{multicols}{2}
2829\lstset{language=CFA,moredelim=**[is][\color{red}]{@}{@},deletedelim=**[is][]{`}{`}}
2830\vspace*{-16pt}
2831\begin{cfa}
2832volatile int go = 0;
2833@monitor@ M {} m;
2834thread T {};
2835void __attribute__((noinline))
2836do_call( M & @mutex@ ) {}
2837void main( T & ) {
2838        while ( go == 0 ) { yield(); }
2839        while ( go == 1 ) { do_call( m ); }
2840}
2841int __attribute__((noinline))
2842do_wait( M & @mutex@ m ) {
2843        go = 1; // continue other thread
2844        BENCH( for ( N ) { @waitfor( do_call, m );@ } )
2845        go = 0; // stop other thread
2846        sout | result`ns;
2847}
2848int main() {
2849        T t;
2850        do_wait( m );
2851}
2852\end{cfa}
2853\captionof{figure}{\CFA external-scheduling benchmark}
2854\label{f:ext-sched}
2855
2856\columnbreak
2857
2858\vspace*{-16pt}
2859\captionof{table}{External-scheduling comparison (nanoseconds)}
2860\label{tab:ext-sched}
2861\begin{tabular}{@{}r*{3}{D{.}{.}{3.2}}@{}}
2862\multicolumn{1}{@{}c}{} & \multicolumn{1}{c}{Median} &\multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
2863\CFA @waitfor@, 1 @monitor@     & 366.7         & 369.5 & 7.52  \\
2864\CFA @waitfor@, 2 @monitor@     & 453.6         & 455.8 & 12.38 \\
2865\CFA @waitfor@, 4 @monitor@     & 671.6         & 672.4 & 14.16 \\
2866\uC @_Accept@                           & 336.0         & 335.8         & 3.22
2867\end{tabular}
2868\end{multicols}
2869
2870
2871\paragraph{Context-Switching}
2872
2873In procedural programming, the cost of a function call is important as modularization (refactoring) increases.
2874(In many cases, a compiler inlines function calls to eliminate this cost.)
2875Similarly, when modularization extends to coroutines/tasks, the time for a context switch becomes a relevant factor.
2876The coroutine test is from resumer to suspender and from suspender to resumer, which is two context switches.
2877The thread test is using yield to enter and return from the runtime kernel, which is two context switches.
2878The difference in performance between coroutine and thread context-switch is the cost of scheduling for threads, whereas coroutines are self-scheduling.
2879Figure~\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}.
2880
2881\begin{multicols}{2}
2882\lstset{language=CFA,moredelim=**[is][\color{red}]{@}{@},deletedelim=**[is][]{`}{`}}
2883\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2884@coroutine@ C {} c;
2885void main( C & ) { for ( ;; ) { @suspend;@ } }
2886int main() { // coroutine test
2887        BENCH( for ( N ) { @resume( c );@ } )
2888        sout | result`ns;
2889}
2890int main() { // task test
2891        BENCH( for ( N ) { @yield();@ } )
2892        sout | result`ns;
2893}
2894\end{cfa}
2895\captionof{figure}{\CFA context-switch benchmark}
2896\label{f:ctx-switch}
2897
2898\columnbreak
2899
2900\vspace*{-16pt}
2901\captionof{table}{Context switch comparison (nanoseconds)}
2902\label{tab:ctx-switch}
2903\begin{tabular}{@{}r*{3}{D{.}{.}{3.2}}@{}}
2904\multicolumn{1}{@{}c}{} & \multicolumn{1}{c}{Median} &\multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
2905C function              & 1.8           & 1.8   & 0             \\
2906\CFA generator  & 2.7           & 2.4   & 0.27  \\
2907\CFA Coroutine  & 37.8          & 37.7  & 0.22  \\
2908\CFA Thread             & 93.6          & 93.8  & 1.46  \\
2909\uC Coroutine   & 52.7          & 52.8  & 0.28  \\
2910\uC Thread              & 93.4          & 93.7  & 1.04  \\
2911Goroutine               & 140.0         & 139.7 & 2.93  \\
2912Java Thread             & 374.0         & 375.8 & 10.38 \\
2913% Qthreads Thread       & 159.5         & 159.3 & 0.71  \\
2914Pthreads Thread & 334.4         & 335.0 & 1.95  \\
2915\end{tabular}
2916\end{multicols}
2917
2918
2919\paragraph{Mutual-Exclusion}
2920
2921Uncontented mutual exclusion, which frequently occurs, is measured by entering/leaving a critical section.
2922For monitors, entering and leaving a monitor function is measured.
2923To 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.
2924Figure~\ref{f:mutex} shows the code for \CFA with all results in Table~\ref{tab:mutex}.
2925Note, the incremental cost of bulk acquire for \CFA, which is largely a fixed cost for small numbers of mutex objects.
2926
2927\begin{multicols}{2}
2928\lstset{language=CFA,moredelim=**[is][\color{red}]{@}{@},deletedelim=**[is][]{`}{`}}
2929\begin{cfa}
2930@monitor@ M {} m1/*, m2, m3, m4*/;
2931void __attribute__((noinline))
2932do_call( M & @mutex m/*, m2, m3, m4*/@ ) {}
2933int main() {
2934        BENCH(
2935                for( N ) do_call( m1/*, m2, m3, m4*/ );
2936        )
2937        sout | result`ns;
2938}
2939\end{cfa}
2940\captionof{figure}{\CFA acquire/release mutex benchmark}
2941\label{f:mutex}
2942
2943\columnbreak
2944
2945\vspace*{-16pt}
2946\captionof{table}{Mutex comparison (nanoseconds)}
2947\label{tab:mutex}
2948\begin{tabular}{@{}r*{3}{D{.}{.}{3.2}}@{}}
2949\multicolumn{1}{@{}c}{} & \multicolumn{1}{c}{Median} &\multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
2950test and test-and-test lock             & 19.1  & 19.0  & 0.36  \\
2951\CFA @mutex@ function, 1 arg.   & 46.6  & 46.8  & 0.86  \\
2952\CFA @mutex@ function, 2 arg.   & 84.1  & 85.3  & 1.86  \\
2953\CFA @mutex@ function, 4 arg.   & 158.6 & 160.7 & 3.07  \\
2954\uC @monitor@ member rtn.               & 54.0  & 53.7  & 0.83  \\
2955Java synchronized method                & 27.0  & 27.1  & 0.25  \\
2956Pthreads Mutex Lock                             & 33.6  & 32.7  & 1.12
2957\end{tabular}
2958\end{multicols}
2959
2960
2961\section{Conclusion}
2962
2963Advanced control-flow will always be difficult, especially when there is temporal ordering and nondeterminism.
2964However, many systems exacerbate the difficulty through their presentation mechanisms.
2965This 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.
2966Eliminated 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.
2967\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@.
2968Extending these mechanisms to handle high-level deadlock-free bulk acquire across both mutual exclusion and synchronization is a unique contribution.
2969The \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.
2970The M:N model is judged to be efficient and provide greater flexibility than a 1:1 threading model.
2971These 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.
2972Performance 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.
2973C 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.
2974
2975
2976\section{Future Work}
2977
2978While control flow in \CFA has a strong start, development is still underway to complete a number of missing features.
2979
2980\paragraph{Flexible Scheduling}
2981\label{futur:sched}
2982
2983An important part of concurrency is scheduling.
2984Different scheduling algorithms can affect performance (both in terms of average and variation).
2985However, no single scheduler is optimal for all workloads and therefore there is value in being able to change the scheduler for given programs.
2986One solution is to offer various tuning options, allowing the scheduler to be adjusted to the requirements of the workload.
2987However, to be truly flexible, a pluggable scheduler is necessary.
2988Currently, 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}.
2989
2990\paragraph{Non-Blocking I/O}
2991\label{futur:nbio}
2992
2993Many modern workloads are not bound by computation but IO operations, a common case being web servers and XaaS~\cite{XaaS} (anything as a service).
2994These types of workloads require significant engineering to amortizing costs of blocking IO-operations.
2995At 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.
2996Current 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.
2997However, these solutions lead to code that is hard to create, read and maintain.
2998A 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.
2999A non-blocking I/O library is currently under development for \CFA.
3000
3001\paragraph{Other Concurrency Tools}
3002\label{futur:tools}
3003
3004While monitors offer flexible and powerful concurrency for \CFA, other concurrency tools are also necessary for a complete multi-paradigm concurrency package.
3005Examples of such tools can include futures and promises~\cite{promises}, executors and actors.
3006These additional features are useful for applications that can be constructed without shared data and direct blocking.
3007As well, new \CFA extensions should make it possible to create a uniform interface for virtually all mutual exclusion, including monitors and low-level locks.
3008
3009\paragraph{Implicit Threading}
3010\label{futur:implcit}
3011
3012Basic 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.
3013This type of concurrency can be achieved both at the language level and at the library level.
3014The canonical example of implicit concurrency is concurrent nested @for@ loops, which are amenable to divide and conquer algorithms~\cite{uC++book}.
3015The \CFA language features should make it possible to develop a reasonable number of implicit concurrency mechanism to solve basic HPC data-concurrency problems.
3016However, implicit concurrency is a restrictive solution with significant limitations, so it can never replace explicit concurrent programming.
3017
3018
3019\section{Acknowledgements}
3020
3021The 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.
3022Funding 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.
3023
3024{%
3025\fontsize{9bp}{12bp}\selectfont%
3026\bibliography{pl,local}
3027}%
3028
3029\end{document}
3030
3031% Local Variables: %
3032% tab-width: 4 %
3033% fill-column: 120 %
3034% compile-command: "make" %
3035% End: %
Note: See TracBrowser for help on using the repository browser.