source: doc/papers/concurrency/Paper.tex @ 1f9a4d0

arm-ehenumforall-pointer-decayjacob/cs343-translationnew-ast-unique-exprpthread-emulationqualifiedEnum
Last change on this file since 1f9a4d0 was 1f9a4d0, checked in by Peter A. Buhr <pabuhr@…>, 2 years ago

changes for SP&E proofs of concurrency paper

  • Property mode set to 100644
File size: 181.0 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, \eg \newterm{abc}.
102% The option parameter provides an index term different from the new term, \eg \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}{for example}
114\newcommand*{\eg}{%
115        \@ifnextchar{,}{\EG}%
116                {\@ifnextchar{:}{\EG}%
117                        {\EG,\xspace}}%
118}}{}%
119\@ifundefined{ie}{
120%\newcommand{\IE}{\abbrevFont{i}\abbrevFont{e}}
121\newcommand{\IE}{that is}
122\newcommand*{\ie}{%
123        \@ifnextchar{,}{\IE}%
124                {\@ifnextchar{:}{\IE}%
125                        {\IE,\xspace}}%
126}}{}%
127\@ifundefined{etc}{
128\newcommand{\ETC}{\abbrevFont{etc}}
129\newcommand*{\etc}{%
130        \@ifnextchar{.}{\ETC}%
131                {\ETC.\xspace}%
132}}{}%
133\@ifundefined{etal}{
134\newcommand{\ETAL}{\abbrevFont{et}~\abbrevFont{al}}
135\newcommand*{\etal}{%
136        \@ifnextchar{.}{\ETAL}%
137                {\ETAL.\xspace}%
138}}{}%
139\@ifundefined{viz}{
140\newcommand{\VIZ}{\abbrevFont{viz}}
141\newcommand*{\viz}{%
142        \@ifnextchar{.}{\VIZ}%
143                {\VIZ.\xspace}%
144}}{}%
145\makeatother
146
147\newenvironment{cquote}
148               {\list{}{\lstset{resetmargins=true,aboveskip=0pt,belowskip=0pt}\topsep=3pt\parsep=0pt\leftmargin=\parindentlnth\rightmargin\leftmargin}%
149                \item\relax}
150               {\endlist}
151
152%\newenvironment{cquote}{%
153%\list{}{\lstset{resetmargins=true,aboveskip=0pt,belowskip=0pt}\topsep=3pt\parsep=0pt\leftmargin=\parindentlnth\rightmargin\leftmargin}%
154%\item\relax%
155%}{%
156%\endlist%
157%}% cquote
158
159% CFA programming language, based on ANSI C (with some gcc additions)
160\lstdefinelanguage{CFA}[ANSI]{C}{
161        morekeywords={
162                _Alignas, _Alignof, __alignof, __alignof__, asm, __asm, __asm__, __attribute, __attribute__,
163                auto, _Bool, catch, catchResume, choose, _Complex, __complex, __complex__, __const, __const__,
164                coroutine, disable, dtype, enable, exception, __extension__, fallthrough, fallthru, finally,
165                __float80, float80, __float128, float128, forall, ftype, generator, _Generic, _Imaginary, __imag, __imag__,
166                inline, __inline, __inline__, __int128, int128, __label__, monitor, mutex, _Noreturn, one_t, or,
167                otype, restrict, resume, __restrict, __restrict__, __signed, __signed__, _Static_assert, suspend, thread,
168                _Thread_local, throw, throwResume, timeout, trait, try, ttype, typeof, __typeof, __typeof__,
169                virtual, __volatile, __volatile__, waitfor, when, with, zero_t},
170        moredirectives={defined,include_next},
171        % replace/adjust listing characters that look bad in sanserif
172        literate={-}{\makebox[1ex][c]{\raisebox{0.5ex}{\rule{0.8ex}{0.1ex}}}}1 {^}{\raisebox{0.6ex}{$\scriptstyle\land\,$}}1
173                {~}{\raisebox{0.3ex}{$\scriptstyle\sim\,$}}1 % {`}{\ttfamily\upshape\hspace*{-0.1ex}`}1
174                {<}{\textrm{\textless}}1 {>}{\textrm{\textgreater}}1
175                {<-}{$\leftarrow$}2 {=>}{$\Rightarrow$}2 {->}{\makebox[1ex][c]{\raisebox{0.5ex}{\rule{0.8ex}{0.075ex}}}\kern-0.2ex{\textrm{\textgreater}}}2,
176}
177
178\lstset{
179language=CFA,
180columns=fullflexible,
181basicstyle=\linespread{0.9}\sf,                                                 % reduce line spacing and use sanserif font
182stringstyle=\tt,                                                                                % use typewriter font
183tabsize=5,                                                                                              % N space tabbing
184xleftmargin=\parindentlnth,                                                             % indent code to paragraph indentation
185%mathescape=true,                                                                               % LaTeX math escape in CFA code $...$
186escapechar=\$,                                                                                  % LaTeX escape in CFA code
187keepspaces=true,                                                                                %
188showstringspaces=false,                                                                 % do not show spaces with cup
189showlines=true,                                                                                 % show blank lines at end of code
190aboveskip=4pt,                                                                                  % spacing above/below code block
191belowskip=3pt,
192moredelim=**[is][\color{red}]{`}{`},
193}% lstset
194
195% uC++ programming language, based on ANSI C++
196\lstdefinelanguage{uC++}[ANSI]{C++}{
197        morekeywords={
198                _Accept, _AcceptReturn, _AcceptWait, _Actor, _At, _CatchResume, _Cormonitor, _Coroutine, _Disable,
199                _Else, _Enable, _Event, _Finally, _Monitor, _Mutex, _Nomutex, _PeriodicTask, _RealTimeTask,
200                _Resume, _Select, _SporadicTask, _Task, _Timeout, _When, _With, _Throw},
201}
202
203% Go programming language: https://github.com/julienc91/listings-golang/blob/master/listings-golang.sty
204\lstdefinelanguage{Golang}{
205        morekeywords=[1]{package,import,func,type,struct,return,defer,panic,recover,select,var,const,iota,},
206        morekeywords=[2]{string,uint,uint8,uint16,uint32,uint64,int,int8,int16,int32,int64,
207                bool,float32,float64,complex64,complex128,byte,rune,uintptr, error,interface},
208        morekeywords=[3]{map,slice,make,new,nil,len,cap,copy,close,true,false,delete,append,real,imag,complex,chan,},
209        morekeywords=[4]{for,break,continue,range,goto,switch,case,fallthrough,if,else,default,},
210        morekeywords=[5]{Println,Printf,Error,},
211        sensitive=true,
212        morecomment=[l]{//},
213        morecomment=[s]{/*}{*/},
214        morestring=[b]',
215        morestring=[b]",
216        morestring=[s]{`}{`},
217        % replace/adjust listing characters that look bad in sanserif
218        literate={-}{\makebox[1ex][c]{\raisebox{0.4ex}{\rule{0.8ex}{0.1ex}}}}1 {^}{\raisebox{0.6ex}{$\scriptstyle\land\,$}}1
219                {~}{\raisebox{0.3ex}{$\scriptstyle\sim\,$}}1 % {`}{\ttfamily\upshape\hspace*{-0.1ex}`}1
220                {<}{\textrm{\textless}}1 {>}{\textrm{\textgreater}}1
221                {<-}{\makebox[2ex][c]{\textrm{\textless}\raisebox{0.5ex}{\rule{0.8ex}{0.075ex}}}}2,
222}
223
224\lstnewenvironment{cfa}[1][]
225{\lstset{#1}}
226{}
227\lstnewenvironment{C++}[1][]                            % use C++ style
228{\lstset{language=C++,moredelim=**[is][\protect\color{red}]{`}{`}}\lstset{#1}}
229{}
230\lstnewenvironment{uC++}[1][]
231{\lstset{language=uC++,moredelim=**[is][\protect\color{red}]{`}{`}}\lstset{#1}}
232{}
233\lstnewenvironment{Go}[1][]
234{\lstset{language=Golang,moredelim=**[is][\protect\color{red}]{`}{`}}\lstset{#1}}
235{}
236\lstnewenvironment{python}[1][]
237{\lstset{language=python,moredelim=**[is][\protect\color{red}]{`}{`}}\lstset{#1}}
238{}
239\lstnewenvironment{java}[1][]
240{\lstset{language=java,moredelim=**[is][\protect\color{red}]{`}{`}}\lstset{#1}}
241{}
242
243% inline code @...@
244\lstMakeShortInline@%
245
246\newcommand{\commenttd}[1]{{\color{red}{Thierry : #1}}}
247
248\let\OLDthebibliography\thebibliography
249\renewcommand\thebibliography[1]{
250  \OLDthebibliography{#1}
251  \setlength{\parskip}{0pt}
252  \setlength{\itemsep}{4pt plus 0.3ex}
253}
254
255\newsavebox{\myboxA}
256\newsavebox{\myboxB}
257\newsavebox{\myboxC}
258\newsavebox{\myboxD}
259
260\title{\texorpdfstring{Advanced Control-flow and Concurrency in \protect\CFA}{Advanced Control-flow in Cforall}}
261
262\author[1]{Thierry Delisle}
263\author[1]{Peter A. Buhr*}
264\authormark{DELISLE \textsc{et al.}}
265
266\address[1]{\orgdiv{Cheriton School of Computer Science}, \orgname{University of Waterloo}, \orgaddress{\state{Waterloo, ON}, \country{Canada}}}
267
268\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}}
269
270% \fundingInfo{Natural Sciences and Engineering Research Council of Canada}
271
272\abstract[Summary]{
273\CFA is a polymorphic, nonobject-oriented, concurrent, backwards compatible extension of the C programming language.
274This paper discusses the design philosophy and implementation of its advanced control-flow and concurrent/parallel features, along with the supporting runtime written in \CFA.
275These features are created from scratch as ISO C has only low-level and/or unimplemented concurrency, so C programmers continue to rely on library approaches like pthreads.
276\CFA introduces modern language-level control-flow mechanisms, like generators, coroutines, user-level threading, and monitors for mutual exclusion and synchronization.
277% Library extension for executors, futures, and actors are built on these basic mechanisms.
278The runtime provides significant programmer simplification and safety by eliminating spurious wakeup and monitor barging.
279The runtime also ensures multiple monitors can be safely acquired in a deadlock-free way, and this feature is fully integrated with all monitor synchronization mechanisms.
280All control-flow features integrate with the \CFA polymorphic type-system and exception handling, while respecting the expectations and style of C programmers.
281Experimental results show comparable performance of the new features with similar mechanisms in other concurrent programming languages.
282}%
283
284\keywords{C \CFA (Cforall) coroutine concurrency generator monitor parallelism runtime thread}
285
286
287\begin{document}
288%\linenumbers                           % comment out to turn off line numbering
289
290\maketitle
291
292
293\section{Introduction}
294
295\CFA~\cite{Moss18,Cforall} is a modern, polymorphic, nonobject-oriented\footnote{
296\CFA has object-oriented features, such as constructors, destructors, and simple trait/interface inheritance.
297% Go interfaces, Rust traits, Swift Protocols, Haskell Type Classes and Java Interfaces.
298% "Trait inheritance" works for me. "Interface inheritance" might also be a good choice, and distinguish clearly from implementation inheritance.
299% You'll want to be a little bit careful with terms like "structural" and "nominal" inheritance as well. CFA has structural inheritance (I think Go as well) -- it's inferred based on the structure of the code.
300% Java, Rust, and Haskell (not sure about Swift) have nominal inheritance, where there needs to be a specific statement that "this type inherits from this type".
301However, functions \emph{cannot} be nested in structures and there is no mechanism to designate a function parameter as a receiver, \lstinline@this@, parameter.},
302, backward-compatible extension of the C programming language.
303In many ways, \CFA is to C as Scala~\cite{Scala} is to Java, providing a vehicle for new typing and control-flow capabilities on top of a highly popular programming language\footnote{
304The TIOBE index~\cite{TIOBE} for May 2020 ranks the top five \emph{popular} programming languages as C 17\%, Java 16\%, Python 9\%, \CC 6\%, and \Csharp 4\% = 52\%, and over the past 30 years, C has always ranked either first or second in popularity.}
305allowing immediate dissemination.
306This paper discusses the design philosophy and implementation of \CFA's advanced control-flow and concurrent/parallel features, along with the supporting runtime written in \CFA.
307
308% The call/return extensions retain state between callee and caller versus losing the callee's state on return;
309% the concurrency extensions allow high-level management of threads.
310
311The \CFA control-flow framework extends ISO \Celeven~\cite{C11} with new call/return and concurrent/parallel control-flow.
312Call/return control-flow with argument and parameter passing appeared in the first programming languages.
313Over the past 50 years, call/return has been augmented with features like static and dynamic call, exceptions (multilevel return) and generators/coroutines (see Section~\ref{s:StatefulFunction}).
314While \CFA has mechanisms for dynamic call (algebraic effects~\cite{Zhang19}) and exceptions\footnote{
315\CFA exception handling will be presented in a separate paper.
316The 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++}}
317, this work only discusses retaining state between calls via generators and coroutines.
318\newterm{Coroutining} was introduced by Conway~\cite{Conway63}, discussed by Knuth~\cite[\S~1.4.2]{Knuth73V1}, implemented in Simula67~\cite{Simula67}, formalized by Marlin~\cite{Marlin80}, and is now popular and appears in old and new programming languages: CLU~\cite{CLU}, \Csharp~\cite{Csharp}, Ruby~\cite{Ruby}, Python~\cite{Python}, JavaScript~\cite{JavaScript}, Lua~\cite{Lua}, \CCtwenty~\cite{C++20Coroutine19}.
319Coroutining is sequential execution requiring direct handoff among coroutines, \ie only the programmer is controlling execution order.
320If coroutines transfer to an internal event-engine for scheduling the next coroutines (as in async-await), the program transitions into the realm of concurrency~\cite[\S~3]{Buhr05a}.
321Coroutines are only a stepping stone toward concurrency where the commonality is that coroutines and threads retain state between calls.
322
323\Celeven and \CCeleven define concurrency~\cite[\S~7.26]{C11}, but it is largely wrappers for a subset of the pthreads library~\cite{Pthreads}.\footnote{Pthreads concurrency is based on simple thread fork and join in a function and mutex or condition locks, which is low-level and error-prone}
324Interestingly, almost a decade after the \Celeven standard, the most recent versions of gcc, clang, and msvc do not support the \Celeven include @threads.h@, indicating no interest in the C11 concurrency approach (possibly because of the recent effort to add concurrency to \CC).
325While 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}, as for \CC.
326In contrast, there has been a renewed interest during the past decade in user-level (M:N, green) threading in old and new programming languages.
327As multicore hardware became available in the 1980/1990s, both user and kernel threading were examined.
328Kernel 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}.
329Libraries like pthreads were developed for C, and the Solaris operating-system switched from user (JDK 1.1~\cite{JDK1.1}) to kernel threads.
330As a result, many languages adopt the 1:1 kernel-threading model, like Java (Scala), Objective-C~\cite{obj-c-book}, \CCeleven~\cite{C11}, C\#~\cite{Csharp} and Rust~\cite{Rust}, with a variety of presentation mechanisms.
331From 2000 onward, several language implementations have championed the M:N user-threading model, like Go~\cite{Go}, Erlang~\cite{Erlang}, Haskell~\cite{Haskell}, D~\cite{D}, and \uC~\cite{uC++,uC++book}, including putting green threads back into Java~\cite{Quasar}, and many user-threading libraries have appeared~\cite{Qthreads,MPC,Marcel}.
332The main argument for user-level threading is that it is lighter weight than kernel threading because locking and context switching do not cross the kernel boundary, so there is less restriction on programming styles that encourages large numbers of threads performing medium-sized work to facilitate load balancing by the runtime~\cite{Verch12}.
333As well, user-threading facilitates a simpler concurrency approach using thread objects that leverage sequential patterns versus events with call-backs~\cite{Adya02,vonBehren03}.
334Finally, performant user-threading implementations, both in time and space, meet or exceed direct kernel-threading implementations, while achieving the programming advantages of high concurrency levels and safety.
335
336A 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, \eg some language features are unsafe in the presence of aggressive sequential optimizations~\cite{Buhr95a,Boehm05}.
337The consequence is that a language must provide sufficient tools to program around safety issues, as inline and library code is compiled as sequential without any explicit concurrent directive.
338One solution is low-level qualifiers and functions, \eg @volatile@ and atomics, allowing \emph{programmers} to explicitly write safe, race-free~\cite{Boehm12} programs.
339A safer solution is high-level language constructs so the \emph{compiler} knows the concurrency boundaries, \ie where mutual exclusion and synchronization are acquired and released, and provide implicit safety at and across these boundaries.
340While the optimization problem is best known with respect to concurrency, it applies to other complex control-flows like exceptions and coroutines.
341As well, language solutions allow matching the language paradigm with the approach, \eg matching the functional paradigm with data-flow programming or the imperative paradigm with thread programming.
342
343Finally, it is important for a language to provide safety over performance \emph{as the default}, allowing careful reduction of safety for performance when necessary.
344Two concurrency violations of this philosophy are \emph{spurious} or \emph{random wakeup}~\cite[\S~9]{Buhr05a}, and \emph{barging}\footnote{
345Barging is competitive succession instead of direct handoff, \ie after a lock is released both arriving and preexisting waiter threads compete to acquire the lock.
346Hence, an arriving thread can temporally \emph{barge} ahead of threads already waiting for an event, which can repeat indefinitely leading to starvation of waiter threads.
347} or signals-as-hints~\cite[\S~8]{Buhr05a}, where one is a consequence of the other, \ie once there is spurious wakeup, barging follows.
348(Author experience teaching concurrency is that students are confused by these semantics.)
349However, spurious wakeup is \emph{not} a foundational concurrency property~\cite[\S~9]{Buhr05a};
350it is a performance design choice.
351We argue removing spurious wakeup and signals-as-hints make concurrent programming simpler and safer as there is less local nondeterminism to manage.
352If barging acquisition is allowed, its specialized performance advantage should be available as an option not the default.
353
354\CFA embraces language extensions for advanced control-flow, user-level threading, and safety as the default.
355We present comparative examples to support our argument that the \CFA control-flow extensions are as expressive and safe as those in other concurrent imperative programming languages, and perform experiments to show the \CFA runtime is competitive with other similar mechanisms.
356The main contributions of this work are:
357\begin{itemize}[topsep=3pt,itemsep=0pt]
358\item
359a set of fundamental execution properties that dictate which language-level control-flow features need to be supported,
360
361\item
362integration of these language-level control-flow features, while respecting the style and expectations of C programmers,
363
364\item
365monitor synchronization without barging, and the ability to safely acquiring multiple monitors in a deadlock-free way, while seamlessly integrating these capabilities with all monitor synchronization mechanisms,
366
367\item
368providing statically type-safe interfaces that integrate with the \CFA polymorphic type-system and other language features,
369
370% \item
371% library extensions for executors, futures, and actors built on the basic mechanisms.
372
373\item
374a runtime system without spurious wake-up and no performance loss,
375
376\item
377a dynamic partitioning mechanism to segregate groups of executing user and kernel threads performing specialized work, \eg web-server or compute engine, or requiring different scheduling, \eg NUMA or real-time.
378
379% \item
380% a nonblocking I/O library
381
382\item
383experimental results showing comparable performance of the \CFA features with similar mechanisms in other languages.
384\end{itemize}
385
386Section~\ref{s:FundamentalExecutionProperties} presents the compositional hierarchy of execution properties directing the design of control-flow features in \CFA.
387Section~\ref{s:StatefulFunction} begins advanced control by introducing sequential functions that retain data and execution state between calls producing constructs @generator@ and @coroutine@.
388Section~\ref{s:Concurrency} begins concurrency, or how to create (fork) and destroy (join) a thread producing the @thread@ construct.
389Section~\ref{s:MutualExclusionSynchronization} discusses the two mechanisms to restricted nondeterminism when controlling shared access to resources, called mutual exclusion, and timing relationships among threads, called synchronization.
390Section~\ref{s:Monitor} shows how both mutual exclusion and synchronization are safely embedded in the @monitor@ and @thread@ constructs.
391Section~\ref{s:CFARuntimeStructure} describes the large-scale mechanism to structure threads and virtual processors (kernel threads).
392Section~\ref{s:Performance} uses microbenchmarks to compare \CFA threading with pthreads, Java 11.0.6, Go 1.12.6, Rust 1.37.0, Python 3.7.6, Node.js v12.18.0, and \uC 7.0.0.
393
394
395\section{Fundamental Execution Properties}
396\label{s:FundamentalExecutionProperties}
397
398The features in a programming language should be composed of a set of fundamental properties rather than an ad hoc collection chosen by the designers.
399To this end, the control-flow features created for \CFA are based on the fundamental properties of any language with function-stack control-flow (see also \uC~\cite[pp.~140-142]{uC++}).
400The fundamental properties are execution state, thread, and mutual-exclusion/synchronization.
401These independent properties can be used to compose different language features, forming a compositional hierarchy, where the combination of all three is the most advanced feature, called a thread.
402While it is possible for a language to only provide threads for composing programs~\cite{Hermes90}, this unnecessarily complicates and makes inefficient solutions to certain classes of problems.
403As is shown, each of the non-rejected composed language features solves a particular set of problems, and hence, has a defensible position in a programming language.
404If a compositional feature is missing, a programmer has too few fundamental properties resulting in a complex and/or inefficient solution.
405
406In detail, the fundamental properties are:
407\begin{description}[leftmargin=\parindent,topsep=3pt,parsep=0pt]
408\item[\newterm{execution state}:]
409It is the state information needed by a control-flow feature to initialize and manage both compute data and execution location(s), and de-initialize.
410For example, calling a function initializes a stack frame including contained objects with constructors, manages local data in blocks and return locations during calls, and de-initializes the frame by running any object destructors and management operations.
411State is retained in fixed-sized aggregate structures (objects) and dynamic-sized stack(s), often allocated in the heap(s) managed by the runtime system.
412The lifetime of state varies with the control-flow feature, where longer life-time and dynamic size provide greater power but also increase usage complexity and cost.
413Control-flow transfers among execution states in multiple ways, such as function call, context switch, asynchronous await, etc.
414Because the programming language determines what constitutes an execution state, implicitly manages this state, and defines movement mechanisms among states, execution state is an elementary property of the semantics of a programming language.
415% An execution-state is related to the notion of a process continuation \cite{Hieb90}.
416
417\item[\newterm{threading}:]
418It is execution of code that occurs independently of other execution, where an individual thread's execution is sequential.
419Multiple threads provide \emph{concurrent execution};
420concurrent execution becomes parallel when run on multiple processing units, \eg hyper-threading, cores, or sockets.
421A programmer needs mechanisms to create, block and unblock, and join with a thread, even if these basic mechanisms are supplied indirectly through high-level features.
422
423\item[\newterm{mutual-exclusion / synchronization (MES)}:]
424It is the concurrency mechanism to perform an action without interruption and establish timing relationships among multiple threads.
425We contented these two properties are independent, \ie mutual exclusion cannot provide synchronization and vice versa without introducing additional threads~\cite[\S~4]{Buhr05a}.
426Limiting MES functionality results in contrived solutions and inefficiency on multicore von Neumann computers where shared memory is a foundational aspect of its design.
427\end{description}
428These properties are fundamental as they cannot be built from existing language features, \eg a basic programming language like C99~\cite{C99} cannot create new control-flow features, concurrency, or provide MES without (atomic) hardware mechanisms.
429
430
431\subsection{Structuring execution properties}
432
433Programming languages seldom present the fundamental execution properties directly to programmers.
434Instead, the properties are packaged into higher-level constructs that encapsulate details and provide safety to these low-level mechanisms.
435Interestingly, language designers often pick and choose among these execution properties proving a varying subset of constructs.
436
437Table~\ref{t:ExecutionPropertyComposition} shows all combinations of the three fundamental execution properties available to language designers.
438(When doing combination case-analysis, not all combinations are meaningful.)
439The combinations of state, thread, and MES compose a hierarchy of control-flow features all of which have appeared in prior programming languages, where each of these languages have found the feature useful.
440To understand the table, it is important to review the basic von Neumann execution requirement of at least one thread and execution state providing some form of call stack.
441For table entries missing these minimal components, the property is borrowed from the invoker (caller).
442Each entry in the table, numbered \textbf{1}--\textbf{12}, is discussed with respect to how the execution properties combine to generate a high-level language construct.
443
444\begin{table}
445\caption{Execution property composition}
446\centering
447\label{t:ExecutionPropertyComposition}
448\renewcommand{\arraystretch}{1.25}
449%\setlength{\tabcolsep}{5pt}
450\vspace*{-5pt}
451\begin{tabular}{c|c||l|l}
452\multicolumn{2}{c||}{Execution properties} & \multicolumn{2}{c}{Mutual exclusion / synchronization} \\
453\hline
454stateful                        & thread        & \multicolumn{1}{c|}{No} & \multicolumn{1}{c}{Yes} \\
455\hline
456\hline
457No                                      & No            & \textbf{1}\ \ \ @struct@                              & \textbf{2}\ \ \ @mutex@ @struct@              \\
458\hline
459Yes (stackless)         & No            & \textbf{3}\ \ \ @generator@                   & \textbf{4}\ \ \ @mutex@ @generator@   \\
460\hline
461Yes (stackful)          & No            & \textbf{5}\ \ \ @coroutine@                   & \textbf{6}\ \ \ @mutex@ @coroutine@   \\
462\hline
463No                                      & Yes           & \textbf{7}\ \ \ {\color{red}rejected} & \textbf{8}\ \ \ {\color{red}rejected} \\
464\hline
465Yes (stackless)         & Yes           & \textbf{9}\ \ \ {\color{red}rejected} & \textbf{10}\ \ \ {\color{red}rejected} \\
466\hline
467Yes (stackful)          & Yes           & \textbf{11}\ \ \ @thread@                             & \textbf{12}\ \ @mutex@ @thread@               \\
468\end{tabular}
469\vspace*{-8pt}
470\end{table}
471
472Case 1 is a structure where access functions borrow local state (stack frame/activation) and thread from the invoker and retain this state across \emph{callees}, \ie function local-variables are retained on the borrowed stack during calls.
473Structures are a foundational mechanism for data organization, and access functions provide interface abstraction and code sharing in all programming languages.
474Case 2 is case 1 with thread safety to a structure's state where access functions provide serialization (mutual exclusion) and scheduling among calling threads (synchronization).
475A @mutex@ structure, often called a \newterm{monitor}, provides a high-level interface for race-free access of shared data in concurrent programming languages.
476Case 3 is case 1 where the structure can implicitly retain execution state and access functions use this execution state to resume/suspend across \emph{callers}, but resume/suspend does not retain a function's local state.
477A stackless structure, often called a \newterm{generator} or \emph{iterator}, is \newterm{stackless} because it still borrows the caller's stack and thread, but the stack is used only to preserve state across its callees not callers.
478Generators provide the first step toward directly solving problems like finite-state machines (FSMs) that retain data and execution state between calls, whereas normal functions restart on each call.
479Case 4 is cases 2 and 3 with thread safety during execution of the generator's access functions.
480A @mutex@ generator extends generators into the concurrent domain.
481Cases 5 and 6 are like cases 3 and 4 where the structure is extended with an implicit separate stack, so only the thread is borrowed by access functions.
482A stackful generator, often called a \newterm{coroutine}, is \newterm{stackful} because resume/suspend now context switch to/from the caller's and coroutine's stack.
483A coroutine extends the state retained between calls beyond the generator's structure to arbitrary call depth in the access functions.
484Cases 7, 8, 9 and 10 are rejected because a new thread must have its own stack, where the thread begins and stack frames are stored for calls, \ie it is unrealistic for a thread to borrow a stack.
485For cases 9 and 10, the stackless frame is not growable, precluding accepting nested calls, making calls, blocking as it requires calls, or preemption as it requires pushing an interrupt frame, all of which compound to require an unknown amount of execution state.
486Hence, if this kind of uninterruptable thread exists, it must execute to completion, \ie computation only, which severely restricts runtime management.
487Cases 11 and 12 are a stackful thread with and without safe access to shared state.
488A thread is the language mechanism to start another thread of control in a program with growable execution state for call/return execution.
489In general, language constructs with more execution properties increase the cost of creation and execution along with complexity of usage.
490
491Given the execution-properties taxonomy, programmers now ask three basic questions: is state necessary across callers and how much, is a separate thread necessary, is thread safety necessary.
492Table~\ref{t:ExecutionPropertyComposition} then suggests the optimal language feature needed for implementing a programming problem.
493The following sections describe how \CFA fills in \emph{all} the nonrejected table entries with language features, while other programming languages may only provide a subset of the table.
494
495
496\subsection{Design requirements}
497
498The following design requirements largely stem from building \CFA on top of C.
499\begin{itemize}[topsep=3pt,parsep=0pt]
500\item
501All communication must be statically type checkable for early detection of errors and efficient code generation.
502This requirement is consistent with the fact that C is a statically typed programming language.
503
504\item
505Direct interaction among language features must be possible allowing any feature to be selected without restricting comm\-unication.
506For example, many concurrent languages do not provide direct communication calls among threads, \ie threads only communicate indirectly through monitors, channels, messages, and/or futures.
507Indirect communication increases the number of objects, consuming more resources, and requires additional synchronization and possibly data transfer.
508
509\item
510All communication is performed using function calls, \ie data are transmitted from argument to parameter and results are returned from function calls.
511Alternative forms of communication, such as call-backs, message passing, channels, or communication ports, step outside of C's normal form of communication.
512
513\item
514All stateful features must follow the same declaration scopes and lifetimes as other language data.
515For C that means at program startup, during block and function activation, and on demand using dynamic allocation.
516
517\item
518MES must be available implicitly in language constructs, \eg Java built-in monitors, as well as explicitly for specialized requirements, \eg @java.util.concurrent@, because requiring programmers to build MES using low-level locks often leads to incorrect programs.
519Furthermore, reducing synchronization scope by encapsulating it within language constructs further reduces errors in concurrent programs.
520
521\item
522Both synchronous and asynchronous communication are needed.
523However, we believe the best way to provide asynchrony, such as call-buffering/chaining and/or returning futures~\cite{multilisp}, is building it from expressive synchronous features.
524
525\item
526Synchronization must be able to control the service order of requests including prioritizing selection from different kinds of outstanding requests, and postponing a request for an unspecified time while continuing to accept new requests.
527Otherwise, certain concurrency problems are difficult, \eg web server, disk scheduling, and the amount of concurrency is inhibited~\cite{Gentleman81}.
528\end{itemize}
529We have satisfied these requirements in \CFA while maintaining backwards compatibility with the huge body of legacy C programs.
530% In contrast, other new programming languages must still access C programs (\eg operating-system service routines), but do so through fragile C interfaces.
531
532
533\subsection{Asynchronous await / call}
534
535Asynchronous await/call is a caller mechanism for structuring programs and/or increasing concurrency, where the caller (client) postpones an action into the future, which is subsequently executed by a callee (server).
536The caller detects the action's completion through a \newterm{future} or \newterm{promise}.
537The benefit is asynchronous caller execution with respect to the callee until future resolution.
538For single-threaded languages like JavaScript, an asynchronous call passes a callee action, which is queued in the event-engine, and continues execution with a promise.
539When the caller needs the promise to be fulfilled, it executes @await@.
540A promise-completion call-back can be part of the callee action or the caller is rescheduled;
541in either case, the call back is executed after the promise is fulfilled.
542While asynchronous calls generate new callee (server) events, we contend this mechanism is insufficient for advanced control-flow mechanisms like generators or coroutines, which are discussed next.
543Specifically, control between caller and callee occurs indirectly through the event-engine precluding direct handoff and cycling among events, and requires complex resolution of a control promise and data.
544Note, @async-await@ is just syntactic-sugar over the event engine so it does not solve these deficiencies.
545For multithreaded languages like Java, the asynchronous call queues a callee action with an executor (server), which subsequently executes the work by a thread in the executor thread-pool.
546The problem is when concurrent work-units need to interact and/or block as this effects the executor by stopping threads.
547While it is possible to extend this approach to support the necessary mechanisms, \eg message passing in Actors, we show monitors and threads provide an equally competitive approach that does not deviate from normal call communication and can be used to build asynchronous call, as is done in Java.
548
549
550\section{Stateful Function}
551\label{s:StatefulFunction}
552
553A \emph{stateful function} has the ability to remember state between calls, where state can be either data or execution, \eg plugin, device driver, FSM.
554A simple technique to retain data state between calls is @static@ declarations within a function, which is often implemented by hoisting the declarations to the global scope but hiding the names within the function using name mangling.
555However, each call starts the function at the top making it difficult to determine the last point of execution in an algorithm, and requiring multiple flag variables and testing to reestablish the continuation point.
556Hence, the next step of generalizing function state is implicitly remembering the return point between calls and reentering the function at this point rather than the top, called \emph{generators}\,/\,\emph{iterators} or \emph{stackless coroutines}.
557For example, a Fibonacci generator retains data and execution state allowing it to remember prior values needed to generate the next value and the location in the algorithm to compute that value.
558The next step of generalization is instantiating the function to allow multiple named instances, \eg multiple Fibonacci generators, where each instance has its own state, and hence, can generate an independent sequence of values.
559Note, a subset of generator state is a function \emph{closure}, \ie the technique of capturing lexical references when returning a nested function.
560A further generalization is adding a stack to a generator's state, called a \emph{coroutine}, so it can suspend outside of itself, \eg call helper functions to arbitrary depth before suspending back to its resumer without unwinding these calls.
561For example, a coroutine iterator for a binary tree can stop the traversal at the visit point (pre, infix, post traversal), return the node value to the caller, and then continue the recursive traversal from the current node on the next call.
562
563There are two styles of activating a stateful function, \emph{asymmetric} or \emph{symmetric}, identified by resume/suspend (no cycles) and resume/resume (cycles).
564These styles \emph{do not} cause incremental stack growth, \eg a million resume/suspend or resume/resume cycles do not remember each cycle just the last resumer for each cycle.
565Selecting between stackless/stackful semantics and asymmetric/symmetric style is a tradeoff between programming requirements, performance, and design, where stackless is faster and smaller using modified call/return between closures, stackful is more general but slower and larger using context switching between distinct stacks, and asymmetric is simpler control-flow than symmetric.
566Additionally, storage management for the closure/stack must be factored into design and performance, especially in unmanaged languages without garbage collection.
567Note, creation cost (closure/stack) is amortized across usage, so activation cost (resume/suspend) is usually the dominant factor.
568
569% The 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.
570% Hence, 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.
571% If 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.
572% If 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.
573% Hence, refactoring a stackless coroutine may require changing it to stackful.
574% A 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.
575% As well, activating a stateful function is \emph{asymmetric} or \emph{symmetric}, identified by resume/suspend (no cycles) and resume/resume (cycles).
576% A fixed closure activated by modified call/return is faster than a variable closure activated by context switching.
577% Additionally, any storage management for the closure (especially in unmanaged languages, \ie no garbage collection) must also be factored into design and performance.
578% Therefore, selecting between stackless and stackful semantics is a tradeoff between programming requirements and performance, where stackless is faster and stackful is more general.
579% nppNote, creation cost is amortized across usage, so activation cost is usually the dominant factor.
580
581For example, Python presents asymmetric generators as a function object, \uC presents symmetric coroutines as a \lstinline[language=C++]|class|-like object, and many languages present threading using function pointers, @pthreads@~\cite{Butenhof97}, \Csharp~\cite{Csharp}, Go~\cite{Go}, and Scala~\cite{Scala}.
582\begin{center}
583\begin{tabular}{@{}l|l|l@{}}
584\multicolumn{1}{@{}c|}{Python asymmetric generator} & \multicolumn{1}{c|}{\uC symmetric coroutine} & \multicolumn{1}{c@{}}{Pthreads thread} \\
585\hline
586\begin{python}
587`def Gen():` $\LstCommentStyle{\color{red}// function}$
588        ... yield val ...
589gen = Gen()
590for i in range( 10 ):
591        print( next( gen ) )
592\end{python}
593&
594\begin{uC++}
595`_Coroutine Cycle {` $\LstCommentStyle{\color{red}// class}$
596        Cycle * p;
597        void main() { p->cycle(); }
598        void cycle() { resume(); }  `};`
599Cycle c1, c2; c1.p=&c2; c2.p=&c1; c1.cycle();
600\end{uC++}
601&
602\begin{cfa}
603void * `rtn`( void * arg ) { ... }
604int i = 3, rc;
605pthread_t t; $\C{// thread id}$
606$\LstCommentStyle{\color{red}// function pointer}$
607rc=pthread_create(&t, `rtn`, (void *)i);
608\end{cfa}
609\end{tabular}
610\end{center}
611\CFA's preferred presentation model for generators/coroutines/threads is a hybrid of functions and classes, giving an object-oriented flavor.
612Essentially, the generator/coroutine/thread function is semantically coupled with a generator/coroutine/thread custom type via the type's name.
613The custom type solves several issues, while accessing the underlying mechanisms used by the custom types is still allowed for flexibility reasons.
614Each custom type is discussed in detail in the following sections.
615
616
617\subsection{Generator}
618
619Stackless generators (Table~\ref{t:ExecutionPropertyComposition} case 3) have the potential to be very small and fast, \ie as small and fast as function call/return for both creation and execution.
620The \CFA goal is to achieve this performance target, possibly at the cost of some semantic complexity.
621A series of different kinds of generators and their implementation demonstrate how this goal is accomplished.\footnote{
622The \CFA operator syntax uses \lstinline|?| to denote operands, which allows precise definitions for pre, post, and infix operators, \eg \lstinline|?++|, \lstinline|++?|, and \lstinline|?+?|, in addition \lstinline|?\{\}| denotes a constructor, as in \lstinline|foo `f` = `\{`...`\}`|, \lstinline|^?\{\}| denotes a destructor, and \lstinline|?()| is \CC function call \lstinline|operator()|.
623Operator \lstinline+|+ is overloaded for printing, like bit-shift \lstinline|<<| in \CC.
624The \CFA \lstinline|with| clause opens an aggregate scope making its fields directly accessible, like Pascal \lstinline|with|, but using parallel semantics;
625multiple aggregates may be opened.
626\CFA has rebindable references \lstinline|int i, & ip = i, j; `&ip = &j;`| and nonrebindable references \lstinline|int i, & `const` ip = i, j; `&ip = &j;` // disallowed|.
627}%
628
629\begin{figure}
630\centering
631\begin{lrbox}{\myboxA}
632\begin{cfa}[aboveskip=0pt,belowskip=0pt]
633typedef struct {
634        int fn1, fn;
635} Fib;
636#define FibCtor { 1, 0 }
637int fib( Fib * f ) {
638
639
640
641
642
643        int fn = f->fn; f->fn = f->fn1;
644                f->fn1 = f->fn + fn;
645        return fn;
646}
647int main() {
648        Fib f1 = FibCtor, f2 = FibCtor;
649        for ( int i = 0; i < 10; i += 1 )
650                printf( "%d %d\n",
651                           fib( &f1 ), fib( &f2 ) );
652}
653\end{cfa}
654\end{lrbox}
655
656\begin{lrbox}{\myboxB}
657\begin{cfa}[aboveskip=0pt,belowskip=0pt]
658`generator` Fib {
659        int fn1, fn;
660};
661
662void `main(Fib & fib)` with(fib) {
663
664
665        [fn1, fn] = [1, 0];
666        for () {
667                `suspend;`
668                [fn1, fn] = [fn, fn + fn1];
669
670        }
671}
672int main() {
673        Fib f1, f2;
674        for ( 10 )
675                sout | `resume( f1 )`.fn
676                         | `resume( f2 )`.fn;
677}
678\end{cfa}
679\end{lrbox}
680
681\begin{lrbox}{\myboxC}
682\begin{cfa}[aboveskip=0pt,belowskip=0pt]
683typedef struct {
684        int `restart`, fn1, fn;
685} Fib;
686#define FibCtor { `0`, 1, 0 }
687Fib * comain( Fib * f ) {
688        `static void * states[] = {&&s0, &&s1};`
689        `goto *states[f->restart];`
690  s0: f->`restart` = 1;
691        for ( ;; ) {
692                return f;
693          s1:; int fn = f->fn + f->fn1;
694                f->fn1 = f->fn; f->fn = fn;
695        }
696}
697int main() {
698        Fib f1 = FibCtor, f2 = FibCtor;
699        for ( int i = 0; i < 10; i += 1 )
700                printf("%d %d\n",comain(&f1)->fn,
701                                 comain(&f2)->fn);
702}
703\end{cfa}
704\end{lrbox}
705
706\subfloat[C]{\label{f:CFibonacci}\usebox\myboxA}
707\hspace{3pt}
708\vrule
709\hspace{3pt}
710\subfloat[\CFA]{\label{f:CFAFibonacciGen}\usebox\myboxB}
711\hspace{3pt}
712\vrule
713\hspace{3pt}
714\subfloat[C generated code for \CFA version]{\label{f:CFibonacciSim}\usebox\myboxC}
715\caption{Fibonacci output asymmetric generator}
716\label{f:FibonacciAsymmetricGenerator}
717
718\bigskip
719
720\begin{lrbox}{\myboxA}
721\begin{cfa}[aboveskip=0pt,belowskip=0pt]
722`generator Fmt` {
723        char ch;
724        int g, b;
725};
726void ?{}( Fmt & fmt ) { `resume(fmt);` } // constructor
727void ^?{}( Fmt & f ) with(f) { $\C[2.25in]{// destructor}$
728        if ( g != 0 || b != 0 ) sout | nl; }
729void `main( Fmt & f )` with(f) {
730        for () { $\C{// until destructor call}$
731                for ( ; g < 5; g += 1 ) { $\C{// groups}$
732                        for ( ; b < 4; b += 1 ) { $\C{// blocks}$
733                                do { `suspend;` $\C{// wait for character}$
734                                while ( ch == '\n' ); // ignore newline
735                                sout | ch;                      $\C{// print character}$
736                        } sout | " ";  $\C{// block separator}$
737                } sout | nl; $\C{// group separator}$
738        }
739}
740int main() {
741        Fmt fmt; $\C{// fmt constructor called}$
742        for () {
743                sin | fmt.ch; $\C{// read into generator}$
744          if ( eof( sin ) ) break;
745                `resume( fmt );`
746        }
747
748} $\C{// fmt destructor called}\CRT$
749\end{cfa}
750\end{lrbox}
751
752\begin{lrbox}{\myboxB}
753\begin{cfa}[aboveskip=0pt,belowskip=0pt]
754typedef struct {
755        int `restart`, g, b;
756        char ch;
757} Fmt;
758void comain( Fmt * f ) {
759        `static void * states[] = {&&s0, &&s1};`
760        `goto *states[f->restart];`
761  s0: f->`restart` = 1;
762        for ( ;; ) {
763                for ( f->g = 0; f->g < 5; f->g += 1 ) {
764                        for ( f->b = 0; f->b < 4; f->b += 1 ) {
765                                do { return;  s1: ;
766                                } while ( f->ch == '\n' );
767                                printf( "%c", f->ch );
768                        } printf( " " );
769                } printf( "\n" );
770        }
771}
772int main() {
773        Fmt fmt = { `0` };  comain( &fmt ); // prime
774        for ( ;; ) {
775                scanf( "%c", &fmt.ch );
776          if ( feof( stdin ) ) break;
777                comain( &fmt );
778        }
779        if ( fmt.g != 0 || fmt.b != 0 ) printf( "\n" );
780}
781\end{cfa}
782\end{lrbox}
783
784\subfloat[\CFA]{\label{f:CFAFormatGen}\usebox\myboxA}
785\hspace{35pt}
786\vrule
787\hspace{3pt}
788\subfloat[C generated code for \CFA version]{\label{f:CFormatGenImpl}\usebox\myboxB}
789\hspace{3pt}
790\caption{Formatter input asymmetric generator}
791\label{f:FormatterAsymmetricGenerator}
792\end{figure}
793
794Figure~\ref{f:FibonacciAsymmetricGenerator} shows an unbounded asymmetric generator for an infinite sequence of Fibonacci numbers written left to right in C, \CFA, and showing the underlying C implementation for the \CFA version.
795This generator is an \emph{output generator}, producing a new result on each resumption.
796To 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.
797An additional requirement is the ability to create an arbitrary number of generators of any kind, \ie retaining one state in global variables is insufficient;
798hence, state is retained in a closure between calls.
799Figure~\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.
800The C version only has the middle execution state because the top execution state is declaration initialization.
801Figure~\ref{f:CFAFibonacciGen} shows the \CFA approach, which also has a manual closure, but replaces the structure with a custom \CFA @generator@ type.
802Each generator type must have a function named \lstinline|main|,
803% \footnote{
804% The name \lstinline|main| has special meaning in C, specifically the function where a program starts execution.
805% Leveraging starting semantics to this name for generator/coroutine/thread is a logical extension.}
806called a \emph{generator main} (leveraging the starting semantics for program @main@ in C), which is connected to the generator type via its single reference parameter.
807The generator main contains @suspend@ statements that suspend execution without ending the generator versus @return@.
808For the Fibonacci generator-main, the top initialization state appears at the start and the middle execution state is denoted by statement @suspend@.
809Any local variables in @main@ \emph{are not retained} between calls;
810hence local variables are only for temporary computations \emph{between} suspends.
811All retained state \emph{must} appear in the generator's type.
812As 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.
813The 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.
814Resuming an ended (returned) generator is undefined.
815Function @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.
816Figure~\ref{f:CFibonacciSim} shows the C implementation of the \CFA asymmetric generator.
817Only one execution-state field, @restart@, is needed to subscript the suspension points in the generator.
818At the start of the generator main, the @static@ declaration, @states@, is initialized to the N suspend points in the generator, where operator @&&@ dereferences or references a label~\cite{gccValueLabels}.
819Next, the computed @goto@ selects the last suspend point and branches to it.
820The cost of setting @restart@ and branching via the computed @goto@ adds very little cost to the suspend and resume calls.
821
822An advantage of the \CFA explicit generator type is the ability to allow multiple type-safe interface functions taking and returning arbitrary types.
823\begin{cfa}
824int ?()( Fib & fib ) { return `resume( fib )`.fn; } $\C[3.9in]{// function-call interface}$
825int ?()( Fib & fib, int N ) { for ( N - 1 ) `fib()`; return `fib()`; } $\C{// add parameter to skip N values}$
826double ?()( Fib & fib ) { return (int)`fib()` / 3.14159; } $\C{// different return type, cast prevents recursive call}$
827Fib f;  int i;  double d;
828i = f();  i = f( 2 );  d = f();                                         $\C{// alternative interfaces}\CRT$
829\end{cfa}
830Now, the generator can be a separately compiled opaque-type only accessed through its interface functions.
831For 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.
832
833\begin{figure}
834%\centering
835\newbox\myboxA
836\begin{lrbox}{\myboxA}
837\begin{python}[aboveskip=0pt,belowskip=0pt]
838def Fib():
839        fn1, fn = 0, 1
840        while True:
841                `yield fn1`
842                fn1, fn = fn, fn1 + fn
843f1 = Fib()
844f2 = Fib()
845for i in range( 10 ):
846        print( next( f1 ), next( f2 ) )
847
848
849
850
851
852
853
854
855
856
857\end{python}
858\end{lrbox}
859
860\newbox\myboxB
861\begin{lrbox}{\myboxB}
862\begin{python}[aboveskip=0pt,belowskip=0pt]
863def Fmt():
864        try:
865                while True:                                             $\C[2.5in]{\# until destructor call}$
866                        for g in range( 5 ):            $\C{\# groups}$
867                                for b in range( 4 ):    $\C{\# blocks}$
868                                        while True:
869                                                ch = (yield)    $\C{\# receive from send}$
870                                                if '\n' not in ch: $\C{\# ignore newline}$
871                                                        break
872                                        print( ch, end='' )     $\C{\# print character}$
873                                print( '  ', end='' )   $\C{\# block separator}$
874                        print()                                         $\C{\# group separator}$
875        except GeneratorExit:                           $\C{\# destructor}$
876                if g != 0 | b != 0:                             $\C{\# special case}$
877                        print()
878fmt = Fmt()
879`next( fmt )`                                                   $\C{\# prime, next prewritten}$
880for i in range( 41 ):
881        `fmt.send( 'a' );`                                      $\C{\# send to yield}$
882\end{python}
883\end{lrbox}
884
885\hspace{30pt}
886\subfloat[Fibonacci]{\label{f:PythonFibonacci}\usebox\myboxA}
887\hspace{3pt}
888\vrule
889\hspace{3pt}
890\subfloat[Formatter]{\label{f:PythonFormatter}\usebox\myboxB}
891\caption{Python generator}
892\label{f:PythonGenerator}
893\end{figure}
894
895Having to manually create the generator closure by moving local-state variables into the generator type is an additional programmer burden (removed by the coroutine in Section~\ref{s:Coroutine}).
896This manual requirement follows from the generality of allowing variable-size local-state, \eg local state with a variable-length array requires dynamic allocation as the array size is unknown at compile time.
897However, dynamic allocation significantly increases the cost of generator creation/destruction and is a showstopper for embedded real-time programming.
898But 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.
899With respect to safety, we believe static analysis can discriminate persistent generator state from temporary generator-main state and raise a compile-time error for temporary usage spanning suspend points.
900Our experience using generators is that the problems have simple data state, including local state, but complex execution state, so the burden of creating the generator type is small.
901As well, C programmers are not afraid of this kind of semantic programming requirement, if it results in very small and fast generators.
902
903Figure~\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.
904\begin{center}
905\tt
906\begin{tabular}{@{}l|l@{}}
907\multicolumn{1}{c|}{\textbf{\textrm{input}}} & \multicolumn{1}{c}{\textbf{\textrm{output}}} \\
908\begin{tabular}[t]{@{}ll@{}}
909abcdefghijklmnopqrstuvwxyz \\
910abcdefghijklmnopqrstuvwxyz
911\end{tabular}
912&
913\begin{tabular}[t]{@{}lllll@{}}
914abcd    & efgh  & ijkl  & mnop  & qrst  \\
915uvwx    & yzab  & cdef  & ghij  & klmn  \\
916opqr    & stuv  & wxyz  &               &
917\end{tabular}
918\end{tabular}
919\end{center}
920The 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.
921The destructor provides a newline, if formatted text ends with a full line.
922Figure~\ref{f:CFormatGenImpl} shows the C implementation of the \CFA input generator with one additional field and the computed @goto@.
923For contrast, Figure~\ref{f:PythonFormatter} shows the equivalent Python format generator with the same properties as the \CFA format generator.
924
925% https://dl-acm-org.proxy.lib.uwaterloo.ca/
926
927An important application for the asymmetric generator is a device-driver, because device drivers are a significant source of operating-system errors: 85\% in Windows XP~\cite[p.~78]{Swift05} and 51.6\% in Linux~\cite[p.~1358,]{Xiao19}. %\cite{Palix11}
928Swift \etal~\cite[p.~86]{Swift05} restructure device drivers using the Extension Procedure Call (XPC) within the kernel via functions @nooks_driver_call@ and @nooks_kernel_call@, which have coroutine properties context switching to separate stacks with explicit hand-off calls;
929however, the calls do not retain execution state, and hence always start from the top.
930The alternative approach for implementing device drivers is using stack-ripping.
931However, Adya \etal~\cite{Adya02} argue against stack ripping in Section 3.2 and suggest a hybrid approach in Section 4 using cooperatively scheduled \emph{fibers}, which is coroutining.
932
933Figure~\ref{f:DeviceDriverGen} shows the generator advantages in implementing a simple network device-driver with the following protocol:
934\begin{center}
935\ldots\, STX \ldots\, message \ldots\, ESC ETX \ldots\, message \ldots\, ETX 2-byte crc \ldots
936\end{center}
937where the network message begins with the control character STX, ends with an ETX, and is followed by a two-byte cyclic-redundancy check.
938Control characters may appear in a message if preceded by an ESC.
939When 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.
940The device driver returns a status code of its current state, and when a complete message is obtained, the operating system reads the message accumulated in the supplied buffer.
941Hence, the device driver is an input/output generator, where the cost of resuming the device-driver generator is the same as call and return, so performance in an operating-system kernel is excellent.
942The key benefits of using a generator are correctness, safety, and maintenance because the execution states are transcribed directly into the programming language rather than table lookup or stack ripping.
943% The conclusion is that FSMs are complex and occur in important domains, so direct generator support is important in a system programming language.
944
945\begin{figure}
946\centering
947\begin{tabular}{@{}l|l@{}}
948\begin{cfa}[aboveskip=0pt,belowskip=0pt]
949enum Status { CONT, MSG, ESTX,
950                                ELNTH, ECRC };
951`generator` Driver {
952        Status status;
953        char byte, * msg; // communication
954        int lnth, sum;      // local state
955        short int crc;
956};
957void ?{}( Driver & d, char * m ) { d.msg = m; }
958Status next( Driver & d, char b ) with( d ) {
959        byte = b; `resume( d );` return status;
960}
961void main( Driver & d ) with( d ) {
962        enum { STX = '\002', ESC = '\033',
963                        ETX = '\003', MaxMsg = 64 };
964  msg: for () { // parse message
965                status = CONT;
966                lnth = 0; sum = 0;
967                while ( byte != STX ) `suspend;`
968          emsg: for () {
969                        `suspend;` // process byte
970\end{cfa}
971&
972\begin{cfa}[aboveskip=0pt,belowskip=0pt]
973                        choose ( byte ) { // switch with implicit break
974                          case STX:
975                                status = ESTX; `suspend;` continue msg;
976                          case ETX:
977                                break emsg;
978                          case ESC:
979                                `suspend;`
980                        }
981                        if ( lnth >= MaxMsg ) { // buffer full ?
982                                status = ELNTH; `suspend;` continue msg; }
983                        msg[lnth++] = byte;
984                        sum += byte;
985                }
986                msg[lnth] = '\0'; // terminate string
987                `suspend;`
988                crc = byte << 8;
989                `suspend;`
990                status = (crc | byte) == sum ? MSG : ECRC;
991                `suspend;`
992        }
993}
994\end{cfa}
995\end{tabular}
996\caption{Device-driver generator for communication protocol}
997\label{f:DeviceDriverGen}
998\end{figure}
999
1000Generators can also have symmetric activation using resume/resume to create control-flow cycles among generators.
1001(The trivial cycle is a generator resuming itself.)
1002This control flow is similar to recursion for functions but without stack growth.
1003Figure~\ref{f:PingPongFullCoroutineSteps} shows the steps for symmetric control-flow using for the ping/pong program in Figure~\ref{f:CFAPingPongGen}.
1004The program starts by creating the generators, @ping@ and @pong@, and then assigns the partners that form the cycle.
1005Constructing 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.
1006(This issue occurs for any cyclic data structure.)
1007% (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.)
1008Once the cycle is formed, the program main resumes one of the generators, @ping@, and the generators can then traverse an arbitrary number of cycles using @resume@ to activate partner generator(s).
1009Terminating 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).
1010Note, the creator and starter may be different, \eg if the creator calls another function that starts the cycle.
1011The starting stack-frame is below the last active generator because the resume/resume cycle does not grow the stack.
1012Also, since local variables are not retained in the generator function, there are no objects with destructors to be called, so the cost is the same as a function return.
1013Destructor cost occurs when the generator instance is deallocated by the creator.
1014
1015\begin{figure}
1016\centering
1017\input{FullCoroutinePhases.pstex_t}
1018\vspace*{-10pt}
1019\caption{Symmetric coroutine steps: Ping / Pong}
1020\label{f:PingPongFullCoroutineSteps}
1021\end{figure}
1022
1023\begin{figure}
1024\centering
1025\begin{lrbox}{\myboxA}
1026\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1027`generator PingPong` {
1028        int N, i;                               // local state
1029        const char * name;
1030        PingPong & partner; // rebindable reference
1031};
1032
1033void `main( PingPong & pp )` with(pp) {
1034
1035
1036        for ( ; i < N; i += 1 ) {
1037                sout | name | i;
1038                `resume( partner );`
1039        }
1040}
1041int main() {
1042        enum { N = 5 };
1043        PingPong ping = {"ping",N,0}, pong = {"pong",N,0};
1044        &ping.partner = &pong;  &pong.partner = &ping;
1045        `resume( ping );`
1046}
1047\end{cfa}
1048\end{lrbox}
1049
1050\begin{lrbox}{\myboxB}
1051\begin{cfa}[escapechar={},aboveskip=0pt,belowskip=0pt]
1052typedef struct PingPong {
1053        int restart, N, i;
1054        const char * name;
1055        struct PingPong * partner;
1056} PingPong;
1057#define PPCtor(name, N) {0, N, 0, name, NULL}
1058void comain( PingPong * pp ) {
1059        static void * states[] = {&&s0, &&s1};
1060        goto *states[pp->restart];
1061  s0: pp->restart = 1;
1062        for ( ; pp->i < pp->N; pp->i += 1 ) {
1063                printf( "%s %d\n", pp->name, pp->i );
1064                asm( "mov  %0,%%rdi" : "=m" (pp->partner) );
1065                asm( "mov  %rdi,%rax" );
1066                asm( "add  $16, %rsp" );
1067                asm( "popq %rbp" );
1068                asm( "jmp  comain" );
1069          s1: ;
1070        }
1071}
1072\end{cfa}
1073\end{lrbox}
1074
1075\subfloat[\CFA symmetric generator]{\label{f:CFAPingPongGen}\usebox\myboxA}
1076\hspace{3pt}
1077\vrule
1078\hspace{3pt}
1079\subfloat[C generator simulation]{\label{f:CPingPongSim}\usebox\myboxB}
1080\hspace{3pt}
1081\caption{Ping-Pong symmetric generator}
1082\label{f:PingPongSymmetricGenerator}
1083\end{figure}
1084
1085Figure~\ref{f:CPingPongSim} shows the C implementation of the \CFA symmetric generator, where there is still only one additional field, @restart@, but @resume@ is more complex because it does a forward rather than backward jump.
1086Before the jump, the parameter for the next call @partner@ is placed into the register used for the first parameter, @rdi@, and the remaining registers are reset for a return.
1087The @jmp comain@ restarts the function but with a different parameter, so the new call's behavior depends on the state of the coroutine type, \ie branch to restart location with different data state.
1088While the semantics of call forward is a tail-call optimization, which compilers perform, the generator state is different on each call rather a common state for a tail-recursive function (\ie the parameter to the function never changes during the forward calls).
1089However, this assembler code depends on what entry code is generated, specifically if there are local variables and the level of optimization.
1090Hence, internal compiler support is necessary for any forward call or backwards return, \eg LLVM has various coroutine support~\cite{CoroutineTS}, and \CFA can leverage this support should it eventually fork @clang@.
1091For this reason, \CFA does not support general symmetric generators at this time, but, it is possible to hand generate any symmetric generators, as in Figure~\ref{f:CPingPongSim}, for proof of concept and performance testing.
1092
1093Finally, part of this generator work was inspired by the recent \CCtwenty coroutine proposal~\cite{C++20Coroutine19}, which uses the general term coroutine to mean generator.
1094Our work provides the same high-performance asymmetric generators as \CCtwenty, and extends their work with symmetric generators.
1095An 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:
1096\begin{cfa}
1097... suspend`{ ... }`;
1098... resume( C )`{ ... }` ...
1099\end{cfa}
1100Since the current generator's stack is released before calling the compound statement, the compound statement can only reference variables in the generator's type.
1101This 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.
1102Hence, this mechanism provides a general and safe handoff of the generator among competing threads.
1103
1104
1105\subsection{Coroutine}
1106\label{s:Coroutine}
1107
1108Stackful coroutines (Table~\ref{t:ExecutionPropertyComposition} case 5) extend generator semantics with an implicit closure and @suspend@ may appear in a helper function called from the coroutine main because of the separate stack.
1109Note, simulating coroutines with stacks of generators, \eg Python with @yield from@ cannot handle symmetric control-flow.
1110Furthermore, all stack components must be of generators, so it is impossible to call a library function passing a generator that yields.
1111Creating a generator copy of the library function maybe impossible because the library function is opaque.
1112
1113A \CFA coroutine is specified by replacing @generator@ with @coroutine@ for the type.
1114Coroutine generality results in higher cost for creation, due to dynamic stack allocation, for execution, due to context switching among stacks, and for terminating, due to possible stack unwinding and dynamic stack deallocation.
1115A series of different kinds of coroutines and their implementations demonstrate how coroutines extend generators.
1116
1117First, 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.
1118Now the coroutine type only contains communication variables between interface functions and the coroutine main.
1119\begin{center}
1120\begin{tabular}{@{}l|l|l|l@{}}
1121\multicolumn{1}{c|}{Fibonacci} & \multicolumn{1}{c|}{Formatter} & \multicolumn{1}{c|}{Device Driver} & \multicolumn{1}{c}{PingPong} \\
1122\hline
1123\begin{cfa}[xleftmargin=0pt]
1124void main( Fib & fib ) ...
1125        `int fn1;`
1126
1127
1128\end{cfa}
1129&
1130\begin{cfa}[xleftmargin=0pt]
1131for ( `g`; 5 ) {
1132        for ( `b`; 4 ) {
1133
1134
1135\end{cfa}
1136&
1137\begin{cfa}[xleftmargin=0pt]
1138status = CONT;
1139`int lnth = 0, sum = 0;`
1140...
1141`short int crc = byte << 8;`
1142\end{cfa}
1143&
1144\begin{cfa}[xleftmargin=0pt]
1145void main( PingPong & pp ) ...
1146        for ( `i`; N ) {
1147
1148
1149\end{cfa}
1150\end{tabular}
1151\end{center}
1152It 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.
1153\begin{cfa}
1154int Crc() {
1155        `suspend;`  short int crc = byte << 8;
1156        `suspend;`  status = (crc | byte) == sum ? MSG : ECRC;
1157        return crc;
1158}
1159\end{cfa}
1160A call to this function is placed at the end of the device driver's coroutine-main.
1161For complex FSMs, refactoring is part of normal program abstraction, especially when code is used in multiple places.
1162Again, this complexity is usually associated with execution state rather than data state.
1163
1164\begin{comment}
1165Figure~\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 @restart@.
1166Like 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.
1167The 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@.
1168The interface function @restart@, takes a Fibonacci instance and context switches to it using @resume@;
1169on restart, the Fibonacci field, @fn@, contains the next value in the sequence, which is returned.
1170The first @resume@ is special because it allocates the coroutine stack and cocalls its coroutine main on that stack;
1171when the coroutine main returns, its stack is deallocated.
1172Hence, @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.
1173Figure~\ref{f:Coroutine1State} shows the coroutine version of the C version in Figure~\ref{f:ExternalState}.
1174Coroutine generators are called \newterm{output coroutines} because values are only returned.
1175
1176\begin{figure}
1177\centering
1178\newbox\myboxA
1179% \begin{lrbox}{\myboxA}
1180% \begin{cfa}[aboveskip=0pt,belowskip=0pt]
1181% `int fn1, fn2, state = 1;`   // single global variables
1182% int fib() {
1183%       int fn;
1184%       `switch ( state )` {  // explicit execution state
1185%         case 1: fn = 0;  fn1 = fn;  state = 2;  break;
1186%         case 2: fn = 1;  fn2 = fn1;  fn1 = fn;  state = 3;  break;
1187%         case 3: fn = fn1 + fn2;  fn2 = fn1;  fn1 = fn;  break;
1188%       }
1189%       return fn;
1190% }
1191% int main() {
1192%
1193%       for ( int i = 0; i < 10; i += 1 ) {
1194%               printf( "%d\n", fib() );
1195%       }
1196% }
1197% \end{cfa}
1198% \end{lrbox}
1199\begin{lrbox}{\myboxA}
1200\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1201#define FibCtor { 0, 1 }
1202typedef struct { int fn1, fn; } Fib;
1203int fib( Fib * f ) {
1204
1205        int ret = f->fn1;
1206        f->fn1 = f->fn;
1207        f->fn = ret + f->fn;
1208        return ret;
1209}
1210
1211
1212
1213int main() {
1214        Fib f1 = FibCtor, f2 = FibCtor;
1215        for ( int i = 0; i < 10; i += 1 ) {
1216                printf( "%d %d\n",
1217                                fib( &f1 ), fib( &f2 ) );
1218        }
1219}
1220\end{cfa}
1221\end{lrbox}
1222
1223\newbox\myboxB
1224\begin{lrbox}{\myboxB}
1225\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1226`coroutine` Fib { int fn1; };
1227void main( Fib & fib ) with( fib ) {
1228        int fn;
1229        [fn1, fn] = [0, 1];
1230        for () {
1231                `suspend;`
1232                [fn1, fn] = [fn, fn1 + fn];
1233        }
1234}
1235int ?()( Fib & fib ) with( fib ) {
1236        return `resume( fib )`.fn1;
1237}
1238int main() {
1239        Fib f1, f2;
1240        for ( 10 ) {
1241                sout | f1() | f2();
1242}
1243
1244
1245\end{cfa}
1246\end{lrbox}
1247
1248\newbox\myboxC
1249\begin{lrbox}{\myboxC}
1250\begin{python}[aboveskip=0pt,belowskip=0pt]
1251
1252def Fib():
1253
1254        fn1, fn = 0, 1
1255        while True:
1256                `yield fn1`
1257                fn1, fn = fn, fn1 + fn
1258
1259
1260// next prewritten
1261
1262
1263f1 = Fib()
1264f2 = Fib()
1265for i in range( 10 ):
1266        print( next( f1 ), next( f2 ) )
1267
1268
1269
1270\end{python}
1271\end{lrbox}
1272
1273\subfloat[C]{\label{f:GlobalVariables}\usebox\myboxA}
1274\hspace{3pt}
1275\vrule
1276\hspace{3pt}
1277\subfloat[\CFA]{\label{f:ExternalState}\usebox\myboxB}
1278\hspace{3pt}
1279\vrule
1280\hspace{3pt}
1281\subfloat[Python]{\label{f:ExternalState}\usebox\myboxC}
1282\caption{Fibonacci generator}
1283\label{f:C-fibonacci}
1284\end{figure}
1285
1286\bigskip
1287
1288\newbox\myboxA
1289\begin{lrbox}{\myboxA}
1290\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1291`coroutine` Fib { int fn; };
1292void main( Fib & fib ) with( fib ) {
1293        fn = 0;  int fn1 = fn; `suspend`;
1294        fn = 1;  int fn2 = fn1;  fn1 = fn; `suspend`;
1295        for () {
1296                fn = fn1 + fn2; fn2 = fn1; fn1 = fn; `suspend`; }
1297}
1298int next( Fib & fib ) with( fib ) { `resume( fib );` return fn; }
1299int main() {
1300        Fib f1, f2;
1301        for ( 10 )
1302                sout | next( f1 ) | next( f2 );
1303}
1304\end{cfa}
1305\end{lrbox}
1306\newbox\myboxB
1307\begin{lrbox}{\myboxB}
1308\begin{python}[aboveskip=0pt,belowskip=0pt]
1309
1310def Fibonacci():
1311        fn = 0; fn1 = fn; `yield fn`  # suspend
1312        fn = 1; fn2 = fn1; fn1 = fn; `yield fn`
1313        while True:
1314                fn = fn1 + fn2; fn2 = fn1; fn1 = fn; `yield fn`
1315
1316
1317f1 = Fibonacci()
1318f2 = Fibonacci()
1319for i in range( 10 ):
1320        print( `next( f1 )`, `next( f2 )` ) # resume
1321
1322\end{python}
1323\end{lrbox}
1324\subfloat[\CFA]{\label{f:Coroutine3States}\usebox\myboxA}
1325\qquad
1326\subfloat[Python]{\label{f:Coroutine1State}\usebox\myboxB}
1327\caption{Fibonacci input coroutine, 3 states, internal variables}
1328\label{f:cfa-fibonacci}
1329\end{figure}
1330\end{comment}
1331
1332\begin{figure}
1333\centering
1334\begin{tabular}{@{}l@{\hspace{2\parindentlnth}}l@{}}
1335\begin{cfa}
1336`coroutine` Prod {
1337        Cons & c;                       $\C[1.5in]{// communication}$
1338        int N, money, receipt;
1339};
1340void main( Prod & prod ) with( prod ) {
1341        for ( i; N ) {          $\C{// 1st resume}\CRT$
1342                int p1 = random( 100 ), p2 = random( 100 );
1343                int status = delivery( c, p1, p2 );
1344                receipt += 1;
1345        }
1346        stop( c );
1347}
1348int payment( Prod & prod, int money ) {
1349        prod.money = money;
1350        `resume( prod );`
1351        return prod.receipt;
1352}
1353void start( Prod & prod, int N, Cons &c ) {
1354        &prod.c = &c;
1355        prod.[N, receipt] = [N, 0];
1356        `resume( prod );`
1357}
1358int main() {
1359        Prod prod;
1360        Cons cons = { prod };
1361        start( prod, 5, cons );
1362}
1363\end{cfa}
1364&
1365\begin{cfa}
1366`coroutine` Cons {
1367        Prod & p;                       $\C[1.5in]{// communication}$
1368        int p1, p2, status;
1369        bool done;
1370};
1371void ?{}( Cons & cons, Prod & p ) {
1372        &cons.p = &p;           $\C{// reassignable reference}$
1373        cons.[status, done ] = [0, false];
1374}
1375void main( Cons & cons ) with( cons ) {
1376        int money = 1, receipt; $\C{// 1st resume}\CRT$
1377        for ( ; ! done; ) {
1378                status += 1;
1379                receipt = payment( p, money );
1380                money += 1;
1381        }
1382}
1383int delivery( Cons & cons, int p1, int p2 ) {
1384        cons.[p1, p2] = [p1, p2];
1385        `resume( cons );`
1386        return cons.status;
1387}
1388void stop( Cons & cons ) {
1389        cons.done = true;
1390        `resume( cons );`
1391}
1392
1393\end{cfa}
1394\end{tabular}
1395\caption{Producer / consumer: resume-resume cycle, bidirectional communication}
1396\label{f:ProdCons}
1397\end{figure}
1398
1399Figure~\ref{f:ProdCons} shows the ping-pong example in Figure~\ref{f:CFAPingPongGen} extended into a producer/consumer symmetric-coroutine performing bidirectional communication.
1400This example is illustrative because both producer and consumer have two interface functions with @resume@s that suspend execution in these interface functions.
1401The 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.
1402The call to @start@ is the first @resume@ of @prod@, which remembers the program main as the starter and creates @prod@'s stack with a frame for @prod@'s coroutine main at the top, and context switches to it.
1403@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's @deliver@ function to transfer the values, and printing the status returned from the consumer.
1404The producer's call to @delivery@ transfers values into the consumer's communication variables, resumes the consumer, and returns the consumer status.
1405Similarly on the first resume, @cons@'s stack is created and initialized, holding local-state variables retained between subsequent activations of the coroutine.
1406The symmetric coroutine cycle forms when the consumer calls the producer's @payment@ function, which resumes the producer in the consumer's delivery function.
1407When the producer calls @delivery@ again, it resumes the consumer in the @payment@ function.
1408Both interface functions then return to their corresponding coroutine-main functions for the next cycle.
1409Figure~\ref{f:ProdConsRuntimeStacks} shows the runtime stacks of the program main, and the coroutine mains for @prod@ and @cons@ during the cycling.
1410As a consequence of a coroutine retaining its last resumer for suspending back, these reverse pointers allow @suspend@ to cycle \emph{backwards} around a symmetric coroutine cycle.
1411
1412\begin{figure}
1413\begin{center}
1414\input{FullProdConsStack.pstex_t}
1415\end{center}
1416\vspace*{-10pt}
1417\caption{Producer / consumer runtime stacks}
1418\label{f:ProdConsRuntimeStacks}
1419\end{figure}
1420
1421Terminating 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.
1422Furthermore, each deallocated coroutine must execute all destructors for objects allocated in the coroutine type \emph{and} allocated on the coroutine's stack at the point of suspension, which can be arbitrarily deep.
1423In the example, termination begins with the producer's loop stopping after N iterations and calling the consumer's @stop@ function, which sets the @done@ flag, resumes the consumer in function @payment@, terminating the call, and the consumer's loop in its coroutine main.
1424% (Not shown is having @prod@ raise a nonlocal @stop@ exception at @cons@ after it finishes generating values and suspend back to @cons@, which catches the @stop@ exception to terminate its loop.)
1425When the consumer's main ends, its stack is already unwound so any stack allocated objects with destructors are finalized.
1426The question now is where does control continue?
1427
1428The na\"{i}ve semantics for coroutine-cycle termination is to context switch to the last resumer, like executing a @suspend@ or @return@ in a generator.
1429However, for coroutines, the last resumer is \emph{not} implicitly below the current stack frame, as for generators, because each coroutine's stack is independent.
1430Unfortunately, it is impossible to determine statically if a coroutine is in a cycle and unrealistic to check dynamically (graph-cycle problem).
1431Hence, a compromise solution is necessary that works for asymmetric (acyclic) and symmetric (cyclic) coroutines.
1432Our solution is to retain a coroutine's starter (first resumer), and context switch back to the starter when the coroutine ends.
1433Hence, the consumer restarts its first resumer, @prod@, in @stop@, and when the producer ends, it restarts its first resumer, program main, in @start@ (see dashed lines from the end of the coroutine mains in Figure~\ref{f:ProdConsRuntimeStacks}).
1434This semantics works well for the most common asymmetric and symmetric coroutine usage patterns.
1435For asymmetric coroutines, it is common for the first resumer (starter) coroutine to be the only resumer;
1436for symmetric coroutines, it is common for the cycle creator to persist for the lifetime of the cycle.
1437For other scenarios, it is always possible to devise a solution with additional programming effort, such as forcing the cycle forward or backward to a safe point before starting termination.
1438
1439Note, the producer/consumer example does not illustrate the full power of the starter semantics because @cons@ always ends first.
1440Assume generator @PingPong@ in Figure~\ref{f:PingPongSymmetricGenerator} is converted to a coroutine.
1441Unlike generators, coroutines have a starter structure with multiple levels, where the program main starts @ping@ and @ping@ starts @pong@.
1442By adjusting $N$ for either @ping@ or @pong@, it is possible to have either finish first.
1443If @pong@ ends first, it resumes its starter @ping@ in its coroutine main, then @ping@ ends and resumes its starter the program main on return;
1444if @ping@ ends first, it resumes its starter the program main on return.
1445Regardless of the cycle complexity, the starter structure always leads back to the program main, but the path can be entered at an arbitrary point.
1446Once back at the program main (creator), coroutines @ping@ and @pong@ are deallocated, running any destructors for objects within the coroutine and possibly deallocating any coroutine stacks for non-terminated coroutines, where stack deallocation implies stack unwinding to find destructors for allocated objects on the stack.
1447Hence, the \CFA termination semantics for the generator and coroutine ensure correct deallocation semantics, regardless of the coroutine's state (terminated or active), like any other aggregate object.
1448
1449
1450\subsection{Generator / coroutine implementation}
1451
1452A significant implementation challenge for generators and 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 the coroutine stack.
1453There are several solutions to this problem, which follow from the object-oriented flavor of adopting custom types.
1454
1455For object-oriented languages, inheritance is used to provide extra fields and code via explicit inheritance:
1456\begin{cfa}[morekeywords={class,inherits}]
1457class myCoroutine inherits baseCoroutine { ... }
1458\end{cfa}
1459% 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.
1460The problem is that some special properties are not handled by existing language semantics, \eg the execution of constructors and 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.
1461Alternatives, such as explicitly starting threads as in Java, are repetitive and forgetting to call start is a common source of errors.
1462An alternative is composition:
1463\begin{cfa}
1464struct myCoroutine {
1465        ... // declaration/communication variables
1466        baseCoroutine dummy; // composition, last declaration
1467}
1468\end{cfa}
1469which also requires an explicit declaration that must be last to ensure correct initialization order.
1470However, there is nothing preventing wrong placement or multiple declarations.
1471
1472\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
1473% and when it is unsafe to perform certain optimizations,
1474and IDEs using simple parsing can find and manipulate types with special properties.
1475The downside of this approach is that it makes custom types a special case in the language.
1476Users wanting to extend custom types or build their own can only do so in ways offered by the language.
1477Furthermore, implementing custom types without language support may display the power of a programming language.
1478\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 (@java.util.concurrent@) monitors.
1479
1480Part 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.
1481\begin{cfa}
1482trait is_coroutine( `dtype` T ) {
1483        void main( T & );
1484        coroutine_desc * get_coroutine( T & );
1485};
1486forall( `dtype` T | is_coroutine(T) ) void $suspend$( T & ), resume( T & );
1487\end{cfa}
1488Note, copying generators, coroutines, and threads is undefined because multiple objects cannot execute on a shared stack and stack copying does not work in unmanaged languages (no garbage collection), like C, because the stack may contain pointers to objects within it that require updating for the copy.
1489The \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 or pointer.
1490The 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 read the coroutine descriptor from its handle.
1491The @main@ function has no return value or additional parameters because the coroutine type allows an arbitrary number of interface functions with arbitrary typed input and output values versus fixed ones.
1492The 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@.
1493
1494The \CFA custom-type @coroutine@ implicitly implements the getter and forward declarations for the coroutine main.
1495\begin{cquote}
1496\begin{tabular}{@{}ccc@{}}
1497\begin{cfa}
1498coroutine MyCor {
1499        int value;
1500
1501};
1502\end{cfa}
1503&
1504{\Large $\Rightarrow$}
1505&
1506\begin{tabular}{@{}ccc@{}}
1507\begin{cfa}
1508struct MyCor {
1509        int value;
1510        coroutine_desc cor;
1511};
1512\end{cfa}
1513&
1514\begin{cfa}
1515static inline coroutine_desc *
1516get_coroutine( MyCor & this ) {
1517        return &this.cor;
1518}
1519\end{cfa}
1520&
1521\begin{cfa}
1522void main( MyCor * this );
1523
1524
1525
1526\end{cfa}
1527\end{tabular}
1528\end{tabular}
1529\end{cquote}
1530The 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.
1531
1532Figure~\ref{f:CoroutineMemoryLayout} shows different memory-layout options for a coroutine (where a thread is similar).
1533The coroutine handle is the @coroutine@ instance containing programmer specified type global and communication variables across interface functions.
1534The coroutine descriptor contains all implicit declarations needed by the runtime, \eg @suspend@/@resume@, and can be part of the coroutine handle or separate.
1535The coroutine stack can appear in a number of locations and be fixed or variable sized.
1536Hence, the coroutine's stack could be a variable-length structure (VLS)
1537% \footnote{
1538% We are examining VLSs, where fields can be variable-sized structures or arrays.
1539% Once allocated, a VLS is fixed sized.}
1540on the allocating stack, provided the allocating stack is large enough.
1541For a VLS stack allocation and deallocation is an inexpensive adjustment of the stack pointer, modulo any stack constructor costs to initial frame setup.
1542For stack allocation in the heap, allocation and deallocation is an expensive allocation, where the heap can be a shared resource, modulo any stack constructor costs.
1543It is also possible to use a split or segmented stack calling convention, available with gcc and clang, allowing a variable-sized stack via a set of connected blocks in the heap.
1544Currently, \CFA supports stack and heap allocated descriptors but only fixed-sized heap allocated stacks.
1545In \CFA debug-mode, the fixed-sized stack is terminated with a write-only page, which catches most stack overflows.
1546Experience teaching concurrency with \uC~\cite{CS343} shows fixed-sized stacks are rarely an issue for students.
1547Split-stack allocation is under development but requires recompilation of legacy code, which is not always possible.
1548
1549\begin{figure}
1550\centering
1551\input{corlayout.pstex_t}
1552\caption{Coroutine memory layout}
1553\label{f:CoroutineMemoryLayout}
1554\end{figure}
1555
1556
1557\section{Concurrency}
1558\label{s:Concurrency}
1559
1560Concurrency is nondeterministic scheduling of independent sequential execution paths (threads), where each thread has its own stack.
1561A single thread with multiple stacks, \ie coroutining, does \emph{not} imply concurrency~\cite[\S~3]{Buhr05a}.
1562Coroutining self-schedule the thread across stacks so execution is deterministic.
1563(It is \emph{impossible} to generate a concurrency error when coroutining.)
1564
1565The 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}.
1566Therefore, a minimal concurrency system requires coroutines \emph{in conjunction with a nondeterministic scheduler}.
1567The resulting execution system now follows a cooperative threading-model~\cite{Adya02,libdill} because context-switching points to the scheduler are known, but the next unblocking point is unknown due to the scheduler.
1568Adding \newterm{preemption} introduces \newterm{non-cooperative} or \newterm{preemptive} scheduling, where context switching points to the scheduler are unknown as they can occur randomly between any two instructions often based on a timer interrupt.
1569Uncertainty gives the illusion of parallelism on a single processor and provides a mechanism to access and increase performance on multiple processors.
1570The reason is that the scheduler and runtime have complete knowledge about resources and how to best utilized them.
1571However, the introduction of unrestricted nondeterminism results in the need for \newterm{mutual exclusion} and \newterm{synchronization}~\cite[\S~4]{Buhr05a}, which restrict nondeterminism for correctness;
1572otherwise, it is impossible to write meaningful concurrent programs.
1573Optimal concurrent performance is often obtained by having as much nondeterminism as mutual exclusion and synchronization correctness allow.
1574
1575A scheduler can also be stackless or stackful.
1576For 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.
1577For 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.
1578The \CFA runtime uses a stackful scheduler for uniformity and security.
1579
1580
1581\subsection{Thread}
1582\label{s:threads}
1583
1584Threading (Table~\ref{t:ExecutionPropertyComposition} case 11) needs the ability to start a thread and wait for its completion, where a common API is @fork@ and @join@.
1585\vspace{4pt}
1586\par\noindent
1587\begin{tabular}{@{}l|l|l@{}}
1588\multicolumn{1}{c|}{\textbf{Java}} & \multicolumn{1}{c|}{\textbf{\Celeven}} & \multicolumn{1}{c}{\textbf{pthreads}} \\
1589\hline
1590\begin{cfa}
1591class MyThread extends Thread {...}
1592mythread t = new MyThread(...);
1593`t.start();` // start
1594// concurrency
1595`t.join();` // wait
1596\end{cfa}
1597&
1598\begin{cfa}
1599class MyThread { ... } // functor
1600MyThread mythread;
1601`thread t( mythread, ... );` // start
1602// concurrency
1603`t.join();` // wait
1604\end{cfa}
1605&
1606\begin{cfa}
1607void * rtn( void * arg ) {...}
1608pthread_t t;  int i = 3;
1609`pthread_create( &t, rtn, (void *)i );` // start
1610// concurrency
1611`pthread_join( t, NULL );` // wait
1612\end{cfa}
1613\end{tabular}
1614\vspace{1pt}
1615\par\noindent
1616\CFA has a simpler approach using a custom @thread@ type and leveraging declaration semantics, allocation and deallocation, where threads implicitly @fork@ after construction and @join@ before destruction.
1617\begin{cfa}
1618thread MyThread {};
1619void main( MyThread & this ) { ... }
1620int main() {
1621        MyThread team`[10]`; $\C[2.5in]{// allocate stack-based threads, implicit start after construction}$
1622        // concurrency
1623} $\C{// deallocate stack-based threads, implicit joins before destruction}$
1624\end{cfa}
1625This semantic ensures a thread is started and stopped exactly once, eliminating some programming error, and scales to multiple threads for basic termination synchronization.
1626For block allocation to arbitrary depth, including recursion, threads are created and destroyed in a lattice structure (tree with top and bottom).
1627Arbitrary topologies are possible using dynamic allocation, allowing threads to outlive their declaration scope, identical to normal dynamic allocation.
1628\begin{cfa}
1629MyThread * factory( int N ) { ... return `anew( N )`; } $\C{// allocate heap-based threads, implicit start after construction}$
1630int main() {
1631        MyThread * team = factory( 10 );
1632        // concurrency
1633        `adelete( team );` $\C{// deallocate heap-based threads, implicit joins before destruction}\CRT$
1634}
1635\end{cfa}
1636
1637Figure~\ref{s:ConcurrentMatrixSummation} shows concurrently adding the rows of a matrix and then totalling the subtotals sequentially, after all the row threads have terminated.
1638The program uses heap-based threads because each thread needs different constructor values.
1639(Python provides a simple iteration mechanism to initialize array elements to different values allowing stack allocation.)
1640The allocation/deallocation pattern appears unusual because allocated objects are immediately deallocated without any intervening code.
1641However, for threads, the deletion provides implicit synchronization, which is the intervening code.
1642% 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.
1643
1644\begin{figure}
1645\begin{cfa}
1646`thread` Adder { int * row, cols, & subtotal; } $\C{// communication variables}$
1647void ?{}( Adder & adder, int row[], int cols, int & subtotal ) {
1648        adder.[ row, cols, &subtotal ] = [ row, cols, &subtotal ];
1649}
1650void main( Adder & adder ) with( adder ) {
1651        subtotal = 0;
1652        for ( c; cols ) { subtotal += row[c]; }
1653}
1654int main() {
1655        const int rows = 10, cols = 1000;
1656        int matrix[rows][cols], subtotals[rows], total = 0;
1657        // read matrix
1658        Adder * adders[rows];
1659        for ( r; rows; ) { $\C{// start threads to sum rows}$
1660                adders[r] = `new( matrix[r], cols, &subtotals[r] );`
1661        }
1662        for ( r; rows ) { $\C{// wait for threads to finish}$
1663                `delete( adders[r] );` $\C{// termination join}$
1664                total += subtotals[r]; $\C{// total subtotal}$
1665        }
1666        sout | total;
1667}
1668\end{cfa}
1669\caption{Concurrent matrix summation}
1670\label{s:ConcurrentMatrixSummation}
1671\end{figure}
1672
1673
1674\subsection{Thread implementation}
1675
1676Threads 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.
1677Like coroutines, and for the same design reasons, \CFA provides a custom @thread@ type and a @trait@ to enforce and restrict the thread-interface functions.
1678\begin{cquote}
1679\begin{tabular}{@{}c@{\hspace{3\parindentlnth}}c@{}}
1680\begin{cfa}
1681thread myThread {
1682        ... // declaration/communication variables
1683};
1684
1685
1686\end{cfa}
1687&
1688\begin{cfa}
1689trait is_thread( `dtype` T ) {
1690        void main( T & );
1691        thread_desc * get_thread( T & );
1692        void ^?{}( T & `mutex` );
1693};
1694\end{cfa}
1695\end{tabular}
1696\end{cquote}
1697Like 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 or pointer.
1698Similarly, the function definitions ensure there is a statically typed @main@ function that is the thread starting point (first stack frame), a mechanism to read the thread descriptor from its handle, and a special destructor to prevent deallocation while the thread is executing.
1699(The qualifier @mutex@ for the destructor parameter is discussed in Section~\ref{s:Monitor}.)
1700The 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;
1701whereas, a thread is scheduling for execution in @main@ immediately after its constructor is run.
1702No 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 and output values.
1703
1704
1705\section{Mutual Exclusion / Synchronization}
1706\label{s:MutualExclusionSynchronization}
1707
1708Unrestricted nondeterminism is meaningless as there is no way to know when a result is completed and safe to access.
1709To 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}.
1710The shared data protected by mutual exclusion is called a \newterm{critical section}~\cite{Dijkstra65}, and the protection can be simple, only 1 thread, or complex, only N kinds of threads, \eg group~\cite{Joung00} or readers/writer~\cite{Courtois71} problems.
1711Without synchronization control in a critical section, an arriving thread can barge ahead of preexisting waiter threads resulting in short/long-term starvation, staleness and freshness problems, and incorrect transfer of data.
1712Preventing or detecting barging is a challenge with low-level locks, but made easier through higher-level constructs.
1713This challenge is often split into two different approaches: barging \emph{avoidance} and \emph{prevention}.
1714Approaches that unconditionally releasing a lock for competing threads to acquire must use barging avoidance with flag/counter variable(s) to force barging threads to wait;
1715approaches that conditionally hold locks during synchronization, \eg baton-passing~\cite{Andrews89}, prevent barging completely.
1716
1717At the lowest level, concurrent control is provided by atomic operations, upon which different kinds of locking mechanisms are constructed, \eg spin locks, semaphores~\cite{Dijkstra68b}, barriers, and path expressions~\cite{Campbell74}.
1718However, for productivity it is always desirable to use the highest-level construct that provides the necessary efficiency~\cite{Hochstein05}.
1719A significant challenge with locks is composability because it takes careful organization for multiple locks to be used while preventing deadlock.
1720Easing composability is another feature higher-level mutual-exclusion mechanisms can offer.
1721Some concurrent systems eliminate mutable shared-state by switching to non-shared 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).
1722However, these approaches introduce a new communication mechanism for concurrency different from the standard communication using function call/return.
1723Hence, a programmer must learn and manipulate two sets of design and programming patterns.
1724While this distinction can be hidden away in library code, effective use of the library still has to take both paradigms into account.
1725In contrast, approaches based on shared-state models more closely resemble the standard call and return programming model, resulting in a single programming paradigm.
1726Finally, a newer approach for restricting non-determinism is transactional memory~\cite{Herlihy93}.
1727While 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~\cite{Cascaval08,Boehm09} to be the main concurrency paradigm for system languages.
1728
1729
1730\section{Monitor}
1731\label{s:Monitor}
1732
1733One of the most natural, elegant, efficient, high-level mechanisms for mutual exclusion and synchronization for shared-memory systems is the \emph{monitor} (Table~\ref{t:ExecutionPropertyComposition} case 2).
1734First 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}.
1735In 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 manually implement a monitor.
1736For these reasons, \CFA selected monitors as the core high-level concurrency construct, upon which higher-level approaches can be easily constructed.
1737
1738Figure~\ref{f:AtomicCounter} compares a \CFA and Java monitor implementing an atomic counter.
1739(Like other concurrent programming languages, \CFA and Java have performant specializations for the basic types using atomic instructions.)
1740A \newterm{monitor} is a set of functions that ensure mutual exclusion when accessing shared state.
1741(Note, in \CFA, @monitor@ is short-hand for @mutex struct@.)
1742More precisely, a monitor is a programming technique that implicitly binds mutual exclusion to static function scope by call and return, as opposed to locks, where mutual exclusion is defined by acquire/release calls, independent of lexical context (analogous to block and heap storage allocation).
1743Restricting acquire and release points eases programming, comprehension, and maintenance, at a slight cost in flexibility and efficiency.
1744As for other special types, \CFA has a custom @monitor@ type.
1745
1746\begin{figure}
1747\centering
1748
1749\begin{lrbox}{\myboxA}
1750\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1751`monitor` Aint { // atomic integer counter
1752        int cnt;
1753};
1754int ++?( Aint & `mutex` this ) with(this) { return ++cnt; }
1755int ?=?( Aint & `mutex` lhs, int rhs ) with(lhs) { cnt = rhs; }
1756int ?=?(int & lhs, Aint & rhs) with(rhs) { lhs = cnt; }
1757
1758int i = 0, j = 0, k = 5;
1759Aint x = { 0 }, y = { 0 }, z = { 5 }; // no mutex
1760++x; ++y; ++z;     // mutex
1761x = 2; y = i; z = k;  // mutex
1762i = x; j = y; k = z;  // no mutex
1763\end{cfa}
1764\end{lrbox}
1765
1766\begin{lrbox}{\myboxB}
1767\begin{java}[aboveskip=0pt,belowskip=0pt]
1768class Aint {
1769    private int cnt;
1770    public Aint( int init ) { cnt = init; }
1771    `synchronized` public int inc() { return ++cnt; }
1772    `synchronized` public void set( int rhs ) {cnt=rhs;}
1773    public int get() { return cnt; }
1774}
1775int i = 0, j = 0, k = 5;
1776Aint x=new Aint(0), y=new Aint(0), z=new Aint(5);
1777x.inc(); y.inc(); z.inc();
1778x.set( 2 ); y.set( i ); z.set( k );
1779i = x.get(); j = y.get(); k = z.get();
1780\end{java}
1781\end{lrbox}
1782
1783\subfloat[\CFA]{\label{f:AtomicCounterCFA}\usebox\myboxA}
1784\hspace{3pt}
1785\vrule
1786\hspace{3pt}
1787\subfloat[Java]{\label{f:AtomicCounterJava}\usebox\myboxB}
1788\caption{Atomic counter}
1789\label{f:AtomicCounter}
1790\end{figure}
1791
1792Like Java, \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.
1793% \begin{cfa}
1794% monitor M { ... } m;
1795% void foo( M & mutex m ) { ... } $\C{// acquire mutual exclusion}$
1796% void bar( M & mutex m ) { $\C{// acquire mutual exclusion}$
1797%       ... `bar( m );` ... `foo( m );` ... $\C{// reacquire mutual exclusion}$
1798% }
1799% \end{cfa}
1800\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.
1801Similar safety is offered by \emph{explicit} opt-in disciplines like \CC RAII versus the monitor \emph{implicit} language-enforced safety guarantee ensuring no programmer usage errors.
1802However, 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;
1803RAII is purely a mutual-exclusion mechanism (see Section~\ref{s:Scheduling}).
1804
1805Both Java and \CFA use a keyword @mutex@/\lstinline[language=java]|synchronized| to designate functions that implicitly acquire/release the monitor lock on call/return providing mutual exclusion to the stared data.
1806Non-designated functions provide no mutual exclusion for read-only access or as an interface to a multi-step protocol requiring several steps of acquiring and releasing the monitor.
1807Monitor objects can be passed through multiple helper functions without acquiring mutual exclusion, until a designated function associated with the object is called.
1808\CFA designated functions are marked by an explicitly parameter-only pointer/reference qualifier @mutex@ (discussed further in Section\ref{s:MutexAcquisition}).
1809Whereas, Java designated members are marked with \lstinline[language=java]|synchronized| that applies to the implicit reference parameter @this@.
1810In the example, the increment and setter operations need mutual exclusion while the read-only getter operation can be nonmutex if reading the implementation is atomic.
1811
1812
1813\subsection{Monitor implementation}
1814
1815For the same design reasons, \CFA provides a custom @monitor@ type and a @trait@ to enforce and restrict the monitor-interface functions.
1816\begin{cquote}
1817\begin{tabular}{@{}c@{\hspace{3\parindentlnth}}c@{}}
1818\begin{cfa}
1819monitor M {
1820        ... // shared data
1821};
1822
1823\end{cfa}
1824&
1825\begin{cfa}
1826trait is_monitor( `dtype` T ) {
1827        monitor_desc * get_monitor( T & );
1828        void ^?{}( T & mutex );
1829};
1830\end{cfa}
1831\end{tabular}
1832\end{cquote}
1833The @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 or pointer.
1834Similarly, the function definitions ensure there is a mechanism to read the monitor descriptor from its handle, and a special destructor to prevent deallocation if a thread is using the shared data.
1835The custom monitor type also inserts any locks needed to implement the mutual exclusion semantics.
1836\CFA relies heavily on traits as an abstraction mechanism, so the @mutex@ qualifier prevents coincidentally matching of a monitor trait with a type that is not a monitor, similar to coincidental inheritance where a shape and playing card can both be drawable.
1837
1838
1839\subsection{Mutex acquisition}
1840\label{s:MutexAcquisition}
1841
1842For object-oriented programming languages, the mutex property applies to one object, the implicit pointer/reference to the monitor type.
1843Because \CFA uses a pointer qualifier, other possibilities exist, \eg:
1844\begin{cfa}
1845monitor M { ... };
1846int f1( M & mutex m ); $\C{// single parameter object}$
1847int f2( M * mutex m ); $\C{// single or multiple parameter object}$
1848int f3( M * mutex m[$\,$] ); $\C{// multiple parameter object}$
1849int f4( stack( M * ) & mutex m ); $\C{// multiple parameters object}$
1850\end{cfa}
1851Function @f1@ has a single object parameter, while functions @f2@ to @f4@ can be a single or multi-element parameter with statically unknown size.
1852Because of the statically unknown size, \CFA only supports a single reference @mutex@ parameter, @f1@.
1853
1854The \CFA @mutex@ qualifier does allow the ability to support multimonitor functions,\footnote{
1855While object-oriented monitors can be extended with a mutex qualifier for multiple-monitor members, no prior example of this feature could be found.}
1856where the number of acquisitions is statically known, called \newterm{bulk acquire}.
1857\CFA guarantees bulk acquisition order is consistent across calls to @mutex@ functions using the same monitors as arguments, so acquiring multiple monitors in a bulk acquire is safe from deadlock.
1858Figure~\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.
1859A \CFA programmer only has to manage when to acquire mutual exclusion;
1860a \CC programmer must select the correct lock and acquisition mechanism from a panoply of locking options.
1861Making good choices for common cases in \CFA simplifies the programming experience and enhances safety.
1862
1863\begin{figure}
1864\centering
1865\begin{lrbox}{\myboxA}
1866\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1867monitor BankAccount {
1868
1869        int balance;
1870} b1 = { 0 }, b2 = { 0 };
1871void deposit( BankAccount & `mutex` b,
1872                        int deposit ) with(b) {
1873        balance += deposit;
1874}
1875void transfer( BankAccount & `mutex` my,
1876        BankAccount & `mutex` your, int me2you ) {
1877        // bulk acquire
1878        deposit( my, -me2you ); // debit
1879        deposit( your, me2you ); // credit
1880}
1881`thread` Person { BankAccount & b1, & b2; };
1882void main( Person & person ) with(person) {
1883        for ( 10_000_000 ) {
1884                if ( random() % 3 ) deposit( b1, 3 );
1885                if ( random() % 3 ) transfer( b1, b2, 7 );
1886        }
1887}
1888int main() {
1889        `Person p1 = { b1, b2 }, p2 = { b2, b1 };`
1890
1891} // wait for threads to complete
1892\end{cfa}
1893\end{lrbox}
1894
1895\begin{lrbox}{\myboxB}
1896\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1897struct BankAccount {
1898        `recursive_mutex m;`
1899        int balance = 0;
1900} b1, b2;
1901void deposit( BankAccount & b, int deposit ) {
1902        `scoped_lock lock( b.m );`
1903        b.balance += deposit;
1904}
1905void transfer( BankAccount & my,
1906                        BankAccount & your, int me2you ) {
1907        `scoped_lock lock( my.m, your.m );` // bulk acquire
1908        deposit( my, -me2you ); // debit
1909        deposit( your, me2you ); // credit
1910}
1911
1912void person( BankAccount & b1, BankAccount & b2 ) {
1913        for ( int i = 0; i < 10$'$000$'$000; i += 1 ) {
1914                if ( random() % 3 ) deposit( b1, 3 );
1915                if ( random() % 3 ) transfer( b1, b2, 7 );
1916        }
1917}
1918int main() {
1919        `thread p1(person, ref(b1), ref(b2)), p2(person, ref(b2), ref(b1));`
1920        `p1.join(); p2.join();`
1921}
1922\end{cfa}
1923\end{lrbox}
1924
1925\subfloat[\CFA]{\label{f:CFABank}\usebox\myboxA}
1926\hspace{3pt}
1927\vrule
1928\hspace{3pt}
1929\subfloat[\CC]{\label{f:C++Bank}\usebox\myboxB}
1930\hspace{3pt}
1931\caption{Bank transfer problem}
1932\label{f:BankTransfer}
1933\end{figure}
1934
1935Users can still force the acquiring order by using or not using @mutex@.
1936\begin{cfa}
1937void foo( M & mutex m1, M & mutex m2 ); $\C{// acquire m1 and m2}$
1938void bar( M & mutex m1, M & m2 ) { $\C{// only acquire m1}$
1939        ... foo( m1, m2 ); ... $\C{// acquire m2}$
1940}
1941void baz( M & m1, M & mutex m2 ) { $\C{// only acquire m2}$
1942        ... foo( m1, m2 ); ... $\C{// acquire m1}$
1943}
1944\end{cfa}
1945The bulk-acquire semantics allow @bar@ or @baz@ to acquire a monitor lock and reacquire it in @foo@.
1946The calls to @bar@ and @baz@ acquired the monitors in opposite order, possibly resulting in deadlock.
1947However, this case is the simplest instance of the \emph{nested-monitor problem}~\cite{Lister77}, where monitors are acquired in sequence versus bulk.
1948Detecting the nested-monitor problem requires dynamic tracking of monitor calls, and dealing with it requires rollback semantics~\cite{Dice10}.
1949\CFA does not deal with this fundamental problem.
1950
1951Finally, like Java, \CFA offers an alternative @mutex@ statement to reduce refactoring and naming.
1952\begin{cquote}
1953\renewcommand{\arraystretch}{0.0}
1954\begin{tabular}{@{}l@{\hspace{3\parindentlnth}}l@{}}
1955\multicolumn{1}{c}{\textbf{\lstinline@mutex@ call}} & \multicolumn{1}{c}{\lstinline@mutex@ \textbf{statement}} \\
1956\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1957monitor M { ... };
1958void foo( M & mutex m1, M & mutex m2 ) {
1959        // critical section
1960}
1961void bar( M & m1, M & m2 ) {
1962        foo( m1, m2 );
1963}
1964\end{cfa}
1965&
1966\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1967
1968void bar( M & m1, M & m2 ) {
1969        mutex( m1, m2 ) {       // remove refactoring and naming
1970                // critical section
1971        }
1972}
1973
1974\end{cfa}
1975\end{tabular}
1976\end{cquote}
1977
1978
1979\subsection{Scheduling}
1980\label{s:Scheduling}
1981
1982% 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.
1983% 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.
1984This section discusses scheduling for waiting threads eligible for monitor entry~\cite{Buhr95b}, \ie which user thread gets the shared resource next.
1985(See Section~\ref{s:RuntimeStructureCluster} for scheduling kernel threads on virtual processors.)
1986While monitor mutual-exclusion provides safe access to its shared data, the data may indicate a thread cannot proceed, \eg a bounded buffer may be full/\-empty so produce/consumer threads must block.
1987Leaving the monitor and retrying (busy waiting) is impractical for high-level programming.
1988
1989Monitors eliminate busy waiting by providing synchronization within the monitor critical-section to schedule threads needing access to the shared data, where threads block versus spin.
1990Synchronization is generally achieved with internal~\cite{Hoare74} or external~\cite[\S~2.9.2]{uC++} scheduling.
1991\newterm{Internal} largely schedules threads located \emph{inside} the monitor and is accomplished using condition variables with signal and wait.
1992\newterm{External} largely schedules threads located \emph{outside} the monitor and is accomplished with the @waitfor@ statement.
1993Note, internal scheduling has a small amount of external scheduling and vice versa, so the naming denotes where the majority of the block threads reside (inside or outside) for scheduling.
1994For complex scheduling, the approaches can be combined, so there are threads waiting inside and outside.
1995
1996\CFA monitors do not allow calling threads to barge ahead of signaled threads via barging prevention, which simplifies synchronization among threads in the monitor and increases correctness.
1997A direct consequence of this semantics is that unblocked waiting threads are not required to recheck the waiting condition, \ie waits are not in a starvation-prone busy-loop as required by the signals-as-hints style with barging.
1998Preventing barging comes directly from Hoare's semantics in the seminal paper on monitors~\cite[p.~550]{Hoare74}.
1999% \begin{cquote}
2000% 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.
2001% It is only in this way that a waiting program has an absolute guarantee that it can acquire the resource just released by the signaling program without any danger that a third program will interpose a monitor entry and seize the resource instead.~\cite[p.~550]{Hoare74}
2002% \end{cquote}
2003Furthermore, \CFA concurrency has no spurious wakeup~\cite[\S~9]{Buhr05a}, which eliminates an implicit self barging.
2004
2005Monitor mutual-exclusion means signaling cannot have the signaller and signaled thread in the monitor simultaneously, so only the signaller or signallee can proceed and the other waits on an implicit urgent list~\cite[p.~551]{Hoare74}.
2006Figure~\ref{f:MonitorScheduling} shows internal and external scheduling for the bounded-buffer examples in Figure~\ref{f:GenericBoundedBuffer}.
2007For internal scheduling in Figure~\ref{f:BBInt}, the @signal@ moves the signallee, front thread of the specified condition queue, to the urgent list (see Figure~\ref{f:MonitorScheduling}) and the signaller continues (solid line).
2008Multiple signals move multiple signallees to urgent until the condition queue is empty.
2009When the signaller exits or waits, a thread is implicitly unblocked from urgent, if available, before unblocking a calling thread to prevent barging.
2010(Java conceptually moves the signaled thread to the calling queue, and hence, allows barging.)
2011Signal is used when the signaller is providing the cooperation needed by the signallee, \eg creating an empty slot in a buffer for a producer, and the signaller immediately exits the monitor to run concurrently consuming the buffer element, and passes control of the monitor to the signaled thread, which can immediately take advantage of the state change.
2012Specifically, the @wait@ function atomically blocks the calling thread and implicitly releases the monitor lock(s) for all monitors in the function's parameter list.
2013Signalling is unconditional because signaling an empty condition queue does nothing.
2014It is common to declare condition queues as monitor fields to prevent shared access, hence no locking is required for access as the queues are protected by the monitor lock.
2015In \CFA, a condition queue can be created and stored independently.
2016
2017\begin{figure}
2018\centering
2019% \subfloat[Scheduling Statements] {
2020% \label{fig:SchedulingStatements}
2021% {\resizebox{0.45\textwidth}{!}{\input{CondSigWait.pstex_t}}}
2022\input{CondSigWait.pstex_t}
2023% }% subfloat
2024% \quad
2025% \subfloat[Bulk acquire monitor] {
2026% \label{fig:BulkMonitor}
2027% {\resizebox{0.45\textwidth}{!}{\input{ext_monitor.pstex_t}}}
2028% }% subfloat
2029\caption{Monitor Scheduling}
2030\label{f:MonitorScheduling}
2031\end{figure}
2032
2033\begin{figure}
2034\centering
2035\newbox\myboxA
2036\begin{lrbox}{\myboxA}
2037\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2038forall( otype T ) { // distribute forall
2039        monitor Buffer {
2040                `condition` full, empty;
2041                int front, back, count;
2042                T elements[10];
2043        };
2044        void ?{}( Buffer(T) & buf ) with(buf) {
2045                front = back = count = 0;
2046        }
2047
2048        void insert(Buffer(T) & mutex buf, T elm) with(buf){
2049                if ( count == 10 ) `wait( empty )`; // full ?
2050                // insert elm into buf
2051                `signal( full )`;
2052        }
2053        T remove( Buffer(T) & mutex buf ) with(buf) {
2054                if ( count == 0 ) `wait( full )`; // empty ?
2055                // remove elm from buf
2056                `signal( empty )`;
2057                return elm;
2058        }
2059}
2060\end{cfa}
2061\end{lrbox}
2062
2063\newbox\myboxB
2064\begin{lrbox}{\myboxB}
2065\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2066forall( otype T ) { // distribute forall
2067        monitor Buffer {
2068
2069                int front, back, count;
2070                T elements[10];
2071        };
2072        void ?{}( Buffer(T) & buf ) with(buf) {
2073                front = back = count = 0;
2074        }
2075        T remove( Buffer(T) & mutex buf ); // forward
2076        void insert(Buffer(T) & mutex buf, T elm) with(buf){
2077                if ( count == 10 ) `waitfor( remove : buf )`;
2078                // insert elm into buf
2079
2080        }
2081        T remove( Buffer(T) & mutex buf ) with(buf) {
2082                if ( count == 0 ) `waitfor( insert : buf )`;
2083                // remove elm from buf
2084
2085                return elm;
2086        }
2087}
2088\end{cfa}
2089\end{lrbox}
2090
2091\subfloat[Internal scheduling]{\label{f:BBInt}\usebox\myboxA}
2092\hspace{1pt}
2093\vrule
2094\hspace{3pt}
2095\subfloat[External scheduling]{\label{f:BBExt}\usebox\myboxB}
2096
2097\caption{Generic bounded buffer}
2098\label{f:GenericBoundedBuffer}
2099\end{figure}
2100
2101The @signal_block@ provides the opposite unblocking order, where the signaller is moved to urgent and the signallee continues and a thread is implicitly unblocked from urgent when the signallee exits or waits (dashed line)~\cite[p.~551]{Hoare74}.
2102Signal block is used when the signallee is providing the cooperation needed by the signaller, \eg if the buffer is removed and a producer hands off an item to a consumer as in Figure~\ref{f:DatingSignalBlock}, so the signaller must wait until the signallee unblocks, provides the cooperation, exits the monitor to run concurrently, and passes control of the monitor to the signaller, which can immediately take advantage of the state change.
2103Using @signal@ or @signal_block@ can be a dynamic decision based on whether the thread providing the cooperation arrives before or after the thread needing the cooperation.
2104
2105For external scheduling in Figure~\ref{f:BBExt}, the internal scheduling is replaced, eliminating condition queues and @signal@/@wait@ (cases where it cannot are discussed shortly), and has existed in the programming language Ada for almost 40 years with variants in other languages~\cite{SR,ConcurrentC++,uC++}.
2106While prior languages use external scheduling solely for thread interaction, \CFA generalizes it to both monitors and threads.
2107External scheduling allows waiting for events from other threads while restricting unrelated events, that would otherwise have to wait on condition queues in the monitor.
2108Scheduling 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.
2109Specifically, a thread calling the monitor is unblocked directly from the calling queue based on function names that can fulfill the cooperation required by the signaller.
2110(The linear search through the calling queue to locate a particular call can be reduced to $O(1)$.)
2111Hence, the @waitfor@ has the same semantics as @signal_block@, where the signallee thread from the calling queue executes before the signaller, which waits on urgent.
2112Now when a producer/consumer detects a full/empty buffer, the necessary cooperation for continuation is specified by indicating the next function call that can occur.
2113For example, a producer detecting a full buffer must have cooperation from a consumer to remove an item so function @remove@ is accepted, which prevents producers from entering the monitor, and after a consumer calls @remove@, the producer waiting on urgent is \emph{implicitly} unblocked because it can now continue its insert operation.
2114Hence, this mechanism is done in terms of control flow, next call, versus in terms of data, channels, as in Go and Rust @select@.
2115While both mechanisms have strengths and weaknesses, \CFA uses the control-flow mechanism to be consistent with other language features.
2116
2117Figure~\ref{f:ReadersWriterLock} shows internal and external scheduling for a readers/writer lock with no barging and threads are serviced in FIFO order to eliminate staleness and freshness among the reader/writer threads.
2118For internal scheduling in Figure~\ref{f:RWInt}, the readers and writers wait on the same condition queue in FIFO order, making it impossible to tell if a waiting thread is a reader or writer.
2119To clawback the kind of thread, a \CFA condition can store user data in the node for a blocking thread at the @wait@, \ie whether the thread is a @READER@ or @WRITER@.
2120An unblocked reader thread checks if the thread at the front of the queue is a reader and unblock it, \ie the readers daisy-chain signal the next group of readers demarcated by the next writer or end of the queue.
2121For external scheduling in Figure~\ref{f:RWExt}, a waiting reader checks if a writer is using the resource, and if so, restricts further calls until the writer exits by calling @EndWrite@.
2122The writer does a similar action for each reader or writer using the resource.
2123Note, no new calls to @StartRead@/@StartWrite@ may occur when waiting for the call to @EndRead@/@EndWrite@.
2124
2125\begin{figure}
2126\centering
2127\newbox\myboxA
2128\begin{lrbox}{\myboxA}
2129\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2130enum RW { READER, WRITER };
2131monitor ReadersWriter {
2132        int rcnt, wcnt; // readers/writer using resource
2133        `condition RWers;`
2134};
2135void ?{}( ReadersWriter & rw ) with(rw) {
2136        rcnt = wcnt = 0;
2137}
2138void EndRead( ReadersWriter & mutex rw ) with(rw) {
2139        rcnt -= 1;
2140        if ( rcnt == 0 ) `signal( RWers )`;
2141}
2142void EndWrite( ReadersWriter & mutex rw ) with(rw) {
2143        wcnt = 0;
2144        `signal( RWers );`
2145}
2146void StartRead( ReadersWriter & mutex rw ) with(rw) {
2147        if ( wcnt !=0 || ! empty( RWers ) )
2148                `wait( RWers, READER )`;
2149        rcnt += 1;
2150        if ( ! empty(RWers) && `front(RWers) == READER` )
2151                `signal( RWers )`;  // daisy-chain signaling
2152}
2153void StartWrite( ReadersWriter & mutex rw ) with(rw) {
2154        if ( wcnt != 0 || rcnt != 0 ) `wait( RWers, WRITER )`;
2155
2156        wcnt = 1;
2157}
2158\end{cfa}
2159\end{lrbox}
2160
2161\newbox\myboxB
2162\begin{lrbox}{\myboxB}
2163\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2164
2165monitor ReadersWriter {
2166        int rcnt, wcnt; // readers/writer using resource
2167
2168};
2169void ?{}( ReadersWriter & rw ) with(rw) {
2170        rcnt = wcnt = 0;
2171}
2172void EndRead( ReadersWriter & mutex rw ) with(rw) {
2173        rcnt -= 1;
2174
2175}
2176void EndWrite( ReadersWriter & mutex rw ) with(rw) {
2177        wcnt = 0;
2178
2179}
2180void StartRead( ReadersWriter & mutex rw ) with(rw) {
2181        if ( wcnt > 0 ) `waitfor( EndWrite : rw );`
2182
2183        rcnt += 1;
2184
2185
2186}
2187void StartWrite( ReadersWriter & mutex rw ) with(rw) {
2188        if ( wcnt > 0 ) `waitfor( EndWrite : rw );`
2189        else while ( rcnt > 0 ) `waitfor( EndRead : rw );`
2190        wcnt = 1;
2191}
2192\end{cfa}
2193\end{lrbox}
2194
2195\subfloat[Internal scheduling]{\label{f:RWInt}\usebox\myboxA}
2196\hspace{1pt}
2197\vrule
2198\hspace{3pt}
2199\subfloat[External scheduling]{\label{f:RWExt}\usebox\myboxB}
2200
2201\caption{Readers / writer lock}
2202\label{f:ReadersWriterLock}
2203\end{figure}
2204
2205Finally, external scheduling requires urgent to be a stack, because the signaller expects to execute immediately after the specified monitor call has exited or waited.
2206Internal scheduling performing multiple signaling results in unblocking from urgent in the reverse order from signaling.
2207It is rare for the unblocking order to be important as an unblocked thread can be time-sliced immediately after leaving the monitor.
2208If the unblocking order is important, multiple signaling can be restructured into daisy-chain signaling, where each thread signals the next thread.
2209Hence, \CFA uses a single urgent stack to correctly handle @waitfor@ and adequately support both forms of signaling.
2210(Advanced @waitfor@ features are discussed in Section~\ref{s:ExtendedWaitfor}.)
2211
2212\begin{figure}
2213\centering
2214\newbox\myboxA
2215\begin{lrbox}{\myboxA}
2216\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2217enum { CCodes = 20 };
2218monitor DS {
2219        int GirlPhNo, BoyPhNo;
2220        condition Girls[CCodes], Boys[CCodes];
2221        `condition exchange;`
2222};
2223int girl( DS & mutex ds, int phNo, int ccode ) {
2224        if ( empty( Boys[ccode] ) ) {
2225                wait( Girls[ccode] );
2226                GirlPhNo = phNo;
2227                `signal( exchange );`
2228        } else {
2229                GirlPhNo = phNo;
2230                `signal( Boys[ccode] );`
2231                `wait( exchange );`
2232        }
2233        return BoyPhNo;
2234}
2235int boy( DS & mutex ds, int phNo, int ccode ) {
2236        // as above with boy/girl interchanged
2237}
2238\end{cfa}
2239\end{lrbox}
2240
2241\newbox\myboxB
2242\begin{lrbox}{\myboxB}
2243\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2244
2245monitor DS {
2246        int GirlPhNo, BoyPhNo;
2247        condition Girls[CCodes], Boys[CCodes];
2248
2249};
2250int girl( DS & mutex ds, int phNo, int ccode ) {
2251        if ( empty( Boys[ccode] ) ) { // no compatible
2252                wait( Girls[ccode] ); // wait for boy
2253                GirlPhNo = phNo; // make phone number available
2254
2255        } else {
2256                GirlPhNo = phNo; // make phone number available
2257                `signal_block( Boys[ccode] );` // restart boy
2258
2259        } // if
2260        return BoyPhNo;
2261}
2262int boy( DS & mutex ds, int phNo, int ccode ) {
2263        // as above with boy/girl interchanged
2264}
2265\end{cfa}
2266\end{lrbox}
2267
2268\subfloat[\lstinline@signal@]{\label{f:DatingSignal}\usebox\myboxA}
2269\qquad
2270\subfloat[\lstinline@signal_block@]{\label{f:DatingSignalBlock}\usebox\myboxB}
2271\caption{Dating service Monitor}
2272\label{f:DatingServiceMonitor}
2273\end{figure}
2274
2275Figure~\ref{f:DatingServiceMonitor} shows a dating service demonstrating nonblocking and blocking signaling.
2276The dating service matches girl and boy threads with matching compatibility codes so they can exchange phone numbers.
2277A thread blocks until an appropriate partner arrives.
2278The complexity is exchanging phone numbers in the monitor because of the mutual-exclusion property.
2279For 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.
2280For signal-block scheduling, the implicit urgent-queue replaces the explicit @exchange@-condition and @signal_block@ puts the finding thread on the urgent stack and unblocks the matcher.
2281Note, 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.
2282This situation shows rechecking the waiting condition and waiting again (signals-as-hints) fails, requiring significant restructured to account for barging.
2283
2284Given external and internal scheduling, what guidelines can a programmer use to select between them?
2285In general, external scheduling is easier to understand and code because only the next logical action (mutex function(s)) is stated, and the monitor implicitly handles all the details.
2286Therefore, there are no condition variables, and hence, no wait and signal, which reduces coding complexity and synchronization errors.
2287If external scheduling is simpler than internal, why not use it all the time?
2288Unfortunately, external scheduling cannot be used if: scheduling depends on parameter value(s) or scheduling must block across an unknown series of calls on a condition variable, \ie internal scheduling.
2289For example, the dating service cannot be written using external scheduling.
2290First, scheduling requires knowledge of calling parameters to make matching decisions and parameters of calling threads are unavailable within the monitor.
2291Specifically, a thread within the monitor cannot examine the @ccode@ of threads waiting on the calling queue to determine if there is a matching partner.
2292(Similarly, if the bounded buffer or readers/writer are restructured with a single interface function with a parameter denoting producer/consumer or reader/write, they cannot be solved with external scheduling.)
2293Second, a scheduling decision may be delayed across an unknown number of calls when there is no immediate match so the thread in the monitor must block on a condition.
2294Specifically, if a thread determines there is no opposite calling thread with the same @ccode@, it must wait an unknown period until a matching thread arrives.
2295For complex synchronization, both external and internal scheduling can be used to take advantage of best of properties of each.
2296
2297Finally, both internal and external scheduling extend to multiple monitors in a natural way.
2298\begin{cquote}
2299\begin{tabular}{@{}l@{\hspace{2\parindentlnth}}l@{}}
2300\begin{cfa}
2301monitor M { `condition e`; ... };
2302void foo( M & mutex m1, M & mutex m2 ) {
2303        ... wait( `e` ); ...   // wait( e, m1, m2 )
2304        ... wait( `e, m1` ); ...
2305        ... wait( `e, m2` ); ...
2306}
2307\end{cfa}
2308&
2309\begin{cfa}
2310void rtn$\(_1\)$( M & mutex m1, M & mutex m2 ); // overload rtn
2311void rtn$\(_2\)$( M & mutex m1 );
2312void bar( M & mutex m1, M & mutex m2 ) {
2313        ... waitfor( `rtn`${\color{red}\(_1\)}$ ); ...       // $\LstCommentStyle{waitfor( rtn\(_1\) : m1, m2 )}$
2314        ... waitfor( `rtn${\color{red}\(_2\)}$ : m1` ); ...
2315}
2316\end{cfa}
2317\end{tabular}
2318\end{cquote}
2319For @wait( e )@, the default semantics is to atomically block the signaller and release all acquired mutex parameters, \ie @wait( e, m1, m2 )@.
2320To override the implicit multimonitor wait, specific mutex parameter(s) can be specified, \eg @wait( e, m1 )@.
2321Wait cannot statically verify the released monitors are the acquired mutex-parameters without disallowing separately compiled helper functions calling @wait@.
2322While \CC supports bulk locking, @wait@ only accepts a single lock for a condition queue, so bulk locking with condition queues is asymmetric.
2323Finally, a signaller,
2324\begin{cfa}
2325void baz( M & mutex m1, M & mutex m2 ) {
2326        ... signal( e ); ...
2327}
2328\end{cfa}
2329must have acquired at least the same locks as the waiting thread signaled from a condition queue to allow the locks to be passed, and hence, prevent barging.
2330
2331Similarly, for @waitfor( rtn )@, the default semantics is to atomically block the acceptor and release all acquired mutex parameters, \ie @waitfor( rtn : m1, m2 )@.
2332To override the implicit multimonitor wait, specific mutex parameter(s) can be specified, \eg @waitfor( rtn : m1 )@.
2333@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 prototype must be accessible.
2334% When an overloaded function appears in an @waitfor@ statement, calls to any function with that name are accepted.
2335% The rationale is that functions with the same name should perform a similar actions, and therefore, all should be eligible to accept a call.
2336Overloaded functions can be disambiguated using a cast
2337\begin{cfa}
2338void rtn( M & mutex m );
2339`int` rtn( M & mutex m );
2340waitfor( (`int` (*)( M & mutex ))rtn : m );
2341\end{cfa}
2342
2343The ability to release a subset of acquired monitors can result in a \newterm{nested monitor}~\cite{Lister77} deadlock (see Section~\ref{s:MutexAcquisition}).
2344\begin{cfa}
2345void foo( M & mutex m1, M & mutex m2 ) {
2346        ... wait( `e, m1` ); ...                                $\C{// release m1, keeping m2 acquired}$
2347void bar( M & mutex m1, M & mutex m2 ) {        $\C{// must acquire m1 and m2}$
2348        ... signal( `e` ); ...
2349\end{cfa}
2350The @wait@ only releases @m1@ so the signaling thread cannot acquire @m1@ and @m2@ to enter @bar@ and @signal@ the condition.
2351While deadlock can occur with multiple/nesting acquisition, this is a consequence of locks, and by extension monitor locking is not perfectly composable.
2352
2353
2354\subsection{\texorpdfstring{Extended \protect\lstinline@waitfor@}{Extended waitfor}}
2355\label{s:ExtendedWaitfor}
2356
2357Figure~\ref{f:ExtendedWaitfor} shows 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.
2358For a @waitfor@ clause to be executed, its @when@ must be true and an outstanding call to its corresponding function(s) must exist.
2359The \emph{conditional-expression} of a @when@ may call a function, but the function must not block or context switch.
2360If there are multiple acceptable mutex calls, selection is prioritized top-to-bottom among the @waitfor@ clauses, whereas some programming languages with similar mechanisms accept nondeterministically for this case, \eg Go \lstinline[morekeywords=select]@select@.
2361If some accept guards are true and there are no outstanding calls to these functions, the acceptor is blocked until a call to one of these functions is made.
2362If there is a @timeout@ clause, it provides an upper bound on waiting.
2363If 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.
2364Hence, the terminating @else@ clause allows a conditional attempt to accept a call without blocking.
2365If both @timeout@ and @else@ clause are present, the @else@ must be conditional, or the @timeout@ is never triggered.
2366% There is also a traditional future wait queue (not shown) (\eg Microsoft @WaitForMultipleObjects@), to wait for a specified number of future elements in the queue.
2367Finally, there is a shorthand for specifying multiple functions using the same set of monitors: @waitfor( f, g, h : m1, m2, m3 )@.
2368
2369\begin{figure}
2370\centering
2371\begin{cfa}
2372`when` ( $\emph{conditional-expression}$ )      $\C{// optional guard}$
2373        waitfor( $\emph{mutex-function-name}$ ) $\emph{statement}$ $\C{// action after call}$
2374`or` `when` ( $\emph{conditional-expression}$ ) $\C{// any number of functions}$
2375        waitfor( $\emph{mutex-function-name}$ ) $\emph{statement}$
2376`or`    ...
2377`when` ( $\emph{conditional-expression}$ ) $\C{// optional guard}$
2378        `timeout` $\emph{statement}$ $\C{// optional terminating timeout clause}$
2379`when` ( $\emph{conditional-expression}$ ) $\C{// optional guard}$
2380        `else`  $\emph{statement}$ $\C{// optional terminating clause}$
2381\end{cfa}
2382\caption{Extended \protect\lstinline@waitfor@}
2383\label{f:ExtendedWaitfor}
2384\end{figure}
2385
2386Note, a group of conditional @waitfor@ clauses is \emph{not} the same as a group of @if@ statements, \eg:
2387\begin{cfa}
2388if ( C1 ) waitfor( mem1 );                       when ( C1 ) waitfor( mem1 );
2389else if ( C2 ) waitfor( mem2 );         or when ( C2 ) waitfor( mem2 );
2390\end{cfa}
2391The left example only accepts @mem1@ if @C1@ is true or only @mem2@ if @C2@ is true.
2392The right example accepts either @mem1@ or @mem2@ if @C1@ and @C2@ are true.
2393Hence, the @waitfor@ has parallel semantics, accepting any true @when@ clause.
2394
2395An interesting use of @waitfor@ is accepting the @mutex@ destructor to know when an object is deallocated, \eg assume the bounded buffer is restructured from a monitor to a thread with the following @main@.
2396\begin{cfa}
2397void main( Buffer(T) & buffer ) with(buffer) {
2398        for () {
2399                `waitfor( ^?{} : buffer )` break;
2400                or when ( count != 20 ) waitfor( insert : buffer ) { ... }
2401                or when ( count != 0 ) waitfor( remove : buffer ) { ... }
2402        }
2403        // clean up
2404}
2405\end{cfa}
2406When the program main deallocates the buffer, it first calls the buffer's destructor, which is accepted, the destructor runs, and the buffer is deallocated.
2407However, the buffer thread cannot continue after the destructor call because the object is gone;
2408hence, clean up in @main@ cannot occur, which means destructors for local objects are not run.
2409To 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.
2410Then, the destructor caller unblocks from urgent to deallocate the object.
2411Accepting the destructor is the idiomatic way in \CFA to terminate a thread performing direct communication.
2412
2413
2414\subsection{Bulk barging prevention}
2415
2416Figure~\ref{f:BulkBargingPrevention} shows \CFA code where bulk acquire adds complexity to the internal-signaling semantics.
2417The complexity begins at the end of the inner @mutex@ statement, where the semantics of internal scheduling need to be extended for multiple monitors.
2418The problem is that bulk acquire is used in the inner @mutex@ statement where one of the monitors is already acquired.
2419When the signaling 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.
2420However, both the signaling and waiting threads W1 and W2 need some subset of monitors @m1@ and @m2@.
2421\begin{cquote}
2422condition c: (order 1) W2(@m2@), W1(@m1@,@m2@)\ \ \ or\ \ \ (order 2) W1(@m1@,@m2@), W2(@m2@) \\
2423S: 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 \\
2424\hspace*{2.75in}$\rightarrow$ rel. @m1@ $\rightarrow$ pass @m1,m2@ unblock W1 (order 1)
2425\end{cquote}
2426
2427\begin{figure}
2428\newbox\myboxA
2429\begin{lrbox}{\myboxA}
2430\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2431monitor M m1, m2;
2432condition c;
2433mutex( m1 ) { // $\LstCommentStyle{\color{red}outer}$
2434        ...
2435        mutex( m1, m2 ) { // $\LstCommentStyle{\color{red}inner}$
2436                ... `signal( c )`; ...
2437                // m1, m2 still acquired
2438        } // $\LstCommentStyle{\color{red}release m2}$
2439        // m1 acquired
2440} // release m1
2441\end{cfa}
2442\end{lrbox}
2443
2444\newbox\myboxB
2445\begin{lrbox}{\myboxB}
2446\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2447
2448
2449mutex( m1 ) {
2450        ...
2451        mutex( m1, m2 ) {
2452                ... `wait( c )`; // release m1, m2
2453                // m1, m2 reacquired
2454        } // $\LstCommentStyle{\color{red}release m2}$
2455        // m1 acquired
2456} // release m1
2457\end{cfa}
2458\end{lrbox}
2459
2460\newbox\myboxC
2461\begin{lrbox}{\myboxC}
2462\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2463
2464
2465mutex( m2 ) {
2466        ... `wait( c )`; // release m2
2467        // m2 reacquired
2468} // $\LstCommentStyle{\color{red}release m2}$
2469
2470
2471
2472
2473\end{cfa}
2474\end{lrbox}
2475
2476\begin{cquote}
2477\subfloat[Signalling Thread (S)]{\label{f:SignallingThread}\usebox\myboxA}
2478\hspace{3\parindentlnth}
2479\subfloat[Waiting Thread (W1)]{\label{f:WaitingThread}\usebox\myboxB}
2480\hspace{2\parindentlnth}
2481\subfloat[Waiting Thread (W2)]{\label{f:OtherWaitingThread}\usebox\myboxC}
2482\end{cquote}
2483\caption{Bulk Barging Prevention}
2484\label{f:BulkBargingPrevention}
2485\end{figure}
2486
2487One 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 behavior of single-monitor scheduling.
2488However, this solution is inefficient if W2 waited first and immediate passed @m2@ when released, while S retains @m1@ until completion of the outer mutex statement.
2489If W1 waited first, the signaller must retain @m1@ amd @m2@ until completion of the outer mutex statement and then pass both to W1.
2490% Furthermore, there is an execution sequence where the signaller always finds waiter W2, and hence, waiter W1 starves.
2491To support these efficient semantics and prevent barging, the implementation maintains a list of monitors acquired for each blocked thread.
2492When a signaller exits or waits in a mutex function or statement, the front waiter on urgent is unblocked if all its monitors are released.
2493Implementing a fast subset check for the necessarily released monitors is important and discussed in the following sections.
2494% 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.
2495
2496
2497\subsection{\texorpdfstring{\protect\lstinline@waitfor@ Implementation}{waitfor Implementation}}
2498\label{s:waitforImplementation}
2499
2500In a statically typed object-oriented programming language, a class has an exhaustive list of members, even when members are added via static inheritance (see Figure~\ref{f:uCinheritance}).
2501Knowing all members at compilation, even separate compilation, allows uniquely numbered them so the accept-statement implementation can use a fast and compact bit mask with $O(1)$ compare.
2502
2503\begin{figure}
2504\centering
2505\begin{lrbox}{\myboxA}
2506\begin{uC++}[aboveskip=0pt,belowskip=0pt]
2507$\emph{translation unit 1}$
2508_Monitor B { // common type in .h file
2509        _Mutex virtual void `f`( ... );
2510        _Mutex virtual void `g`( ... );
2511        _Mutex virtual void w1( ... ) { ... _Accept(`f`, `g`); ... }
2512};
2513$\emph{translation unit 2}$
2514// include B
2515_Monitor D : public B { // inherit
2516        _Mutex void `h`( ... ); // add
2517        _Mutex void w2( ... ) { ... _Accept(`f`, `h`); ... }
2518};
2519\end{uC++}
2520\end{lrbox}
2521
2522\begin{lrbox}{\myboxB}
2523\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2524$\emph{translation unit 1}$
2525monitor M { ... }; // common type in .h file
2526void `f`( M & mutex m, ... );
2527void `g`( M & mutex m, ... );
2528void w1( M & mutex m, ... ) { ... waitfor(`f`, `g` : m); ... }
2529
2530$\emph{translation unit 2}$
2531// include M
2532extern void `f`( M & mutex m, ... ); // import f but not g
2533void `h`( M & mutex m ); // add
2534void w2( M & mutex m, ... ) { ... waitfor(`f`, `h` : m); ... }
2535
2536\end{cfa}
2537\end{lrbox}
2538
2539\subfloat[\uC]{\label{f:uCinheritance}\usebox\myboxA}
2540\hspace{3pt}
2541\vrule
2542\hspace{3pt}
2543\subfloat[\CFA]{\label{f:CFinheritance}\usebox\myboxB}
2544\caption{Member / function visibility}
2545\label{f:MemberFunctionVisibility}
2546\end{figure}
2547
2548However, the @waitfor@ statement in translation unit 2 (see Figure~\ref{f:CFinheritance}) cannot see function @g@ in translation unit 1 precluding a unique numbering for a bit-mask because the monitor type only carries the protected shared data.
2549(A possible way to construct a dense mapping is at link or load-time.)
2550Hence, function pointers are used to identify the functions listed in the @waitfor@ statement, stored in a variable-sized array.
2551Then, the same implementation approach used for the urgent stack (see Section~\ref{s:Scheduling}) is used for the calling queue.
2552Each caller has a list of monitors acquired, and the @waitfor@ statement performs a short linear search matching functions in the @waitfor@ list with called functions, and then verifying the associated mutex locks can be transferred.
2553
2554
2555\subsection{Multimonitor scheduling}
2556\label{s:Multi-MonitorScheduling}
2557
2558External scheduling, like internal scheduling, becomes significantly more complex for multimonitor semantics.
2559Even in the simplest case, new semantics need to be established.
2560\begin{cfa}
2561monitor M { ... };
2562void f( M & mutex m1 );
2563void g( M & mutex m1, M & mutex m2 ) { `waitfor( f );` } $\C{// pass m1 or m2 to f?}$
2564\end{cfa}
2565The solution is for the programmer to disambiguate:
2566\begin{cfa}
2567waitfor( f : `m2` ); $\C{// wait for call to f with argument m2}$
2568\end{cfa}
2569Both 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@.
2570This behavior can be extended to the multimonitor @waitfor@ statement.
2571\begin{cfa}
2572monitor M { ... };
2573void f( M & mutex m1, M & mutex m2 );
2574void g( M & mutex m1, M & mutex m2 ) { waitfor( f : `m1, m2` ); $\C{// wait for call to f with arguments m1 and m2}$
2575\end{cfa}
2576Again, the set of monitors passed to the @waitfor@ statement must be entirely contained in the set of monitors already acquired by the accepting function.
2577% Also, the order of the monitors in a @waitfor@ statement must match the order of the mutex parameters.
2578
2579Figure~\ref{f:UnmatchedMutexSets} shows internal and external scheduling with multiple monitors that must match exactly with a signaling or accepting thread, \ie partial matching results in waiting.
2580In both cases, the set of monitors is disjoint so unblocking is impossible.
2581
2582\begin{figure}
2583\centering
2584\begin{lrbox}{\myboxA}
2585\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2586monitor M1 {} m11, m12;
2587monitor M2 {} m2;
2588condition c;
2589void f( M1 & mutex m1, M2 & mutex m2 ) {
2590        signal( c );
2591}
2592void g( M1 & mutex m1, M2 & mutex m2 ) {
2593        wait( c );
2594}
2595g( `m11`, m2 ); // block on wait
2596f( `m12`, m2 ); // cannot fulfil
2597\end{cfa}
2598\end{lrbox}
2599
2600\begin{lrbox}{\myboxB}
2601\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2602monitor M1 {} m11, m12;
2603monitor M2 {} m2;
2604
2605void f( M1 & mutex m1, M2 & mutex m2 ) {
2606
2607}
2608void g( M1 & mutex m1, M2 & mutex m2 ) {
2609        waitfor( f : m1, m2 );
2610}
2611g( `m11`, m2 ); // block on accept
2612f( `m12`, m2 ); // cannot fulfil
2613\end{cfa}
2614\end{lrbox}
2615\subfloat[Internal scheduling]{\label{f:InternalScheduling}\usebox\myboxA}
2616\hspace{3pt}
2617\vrule
2618\hspace{3pt}
2619\subfloat[External scheduling]{\label{f:ExternalScheduling}\usebox\myboxB}
2620\caption{Unmatched \protect\lstinline@mutex@ sets}
2621\label{f:UnmatchedMutexSets}
2622\end{figure}
2623
2624\begin{figure}
2625\centering
2626\begin{lrbox}{\myboxA}
2627\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2628
2629struct Msg { int i, j; };
2630mutex thread GoRtn { int i;  float f;  Msg m; };
2631void mem1( GoRtn & mutex gortn, int i ) { gortn.i = i; }
2632void mem2( GoRtn & mutex gortn, float f ) { gortn.f = f; }
2633void mem3( GoRtn & mutex gortn, Msg m ) { gortn.m = m; }
2634void ^?{}( GoRtn & mutex ) {}
2635
2636void main( GoRtn & mutex gortn ) with(gortn) { // thread starts
2637
2638        for () {
2639
2640                `waitfor( mem1 : gortn )` sout | i;  // wait for calls
2641                or `waitfor( mem2 : gortn )` sout | f;
2642                or `waitfor( mem3 : gortn )` sout | m.i | m.j;
2643                or `waitfor( ^?{} : gortn )` break; // low priority
2644
2645        }
2646
2647}
2648int main() {
2649        GoRtn gortn; $\C[2.0in]{// start thread}$
2650        `mem1( gortn, 0 );` $\C{// different calls}\CRT$
2651        `mem2( gortn, 2.5 );`
2652        `mem3( gortn, (Msg){1, 2} );`
2653
2654
2655} // wait for completion
2656\end{cfa}
2657\end{lrbox}
2658
2659\begin{lrbox}{\myboxB}
2660\begin{Go}[aboveskip=0pt,belowskip=0pt]
2661func main() {
2662        type Msg struct{ i, j int }
2663
2664        ch1 := make( chan int )
2665        ch2 := make( chan float32 )
2666        ch3 := make( chan Msg )
2667        hand := make( chan string )
2668        shake := make( chan string )
2669        gortn := func() { $\C[1.5in]{// thread starts}$
2670                var i int;  var f float32;  var m Msg
2671                L: for {
2672                        select { $\C{// wait for messages}$
2673                          case `i = <- ch1`: fmt.Println( i )
2674                          case `f = <- ch2`: fmt.Println( f )
2675                          case `m = <- ch3`: fmt.Println( m )
2676                          case `<- hand`: break L $\C{// sentinel}$
2677                        }
2678                }
2679                `shake <- "SHAKE"` $\C{// completion}$
2680        }
2681
2682        go gortn() $\C{// start thread}$
2683        `ch1 <- 0` $\C{// different messages}$
2684        `ch2 <- 2.5`
2685        `ch3 <- Msg{1, 2}`
2686        `hand <- "HAND"` $\C{// sentinel value}$
2687        `<- shake` $\C{// wait for completion}\CRT$
2688}
2689\end{Go}
2690\end{lrbox}
2691
2692\subfloat[\CFA]{\label{f:CFAwaitfor}\usebox\myboxA}
2693\hspace{3pt}
2694\vrule
2695\hspace{3pt}
2696\subfloat[Go]{\label{f:Gochannel}\usebox\myboxB}
2697\caption{Direct versus indirect communication}
2698\label{f:DirectCommunicationComparison}
2699
2700\medskip
2701
2702\begin{cfa}
2703mutex thread DatingService {
2704        condition Girls[CompCodes], Boys[CompCodes];
2705        int girlPhoneNo, boyPhoneNo, ccode;
2706};
2707int girl( DatingService & mutex ds, int phoneno, int code ) with( ds ) {
2708        girlPhoneNo = phoneno;  ccode = code;
2709        `wait( Girls[ccode] );`                                                         $\C{// wait for boy}$
2710        girlPhoneNo = phoneno;  return boyPhoneNo;
2711}
2712int boy( DatingService & mutex ds, int phoneno, int code ) with( ds ) {
2713        boyPhoneNo = phoneno;  ccode = code;
2714        `wait( Boys[ccode] );`                                                          $\C{// wait for girl}$
2715        boyPhoneNo = phoneno;  return girlPhoneNo;
2716}
2717void main( DatingService & ds ) with( ds ) {                    $\C{// thread starts, ds defaults to mutex}$
2718        for () {
2719                waitfor( ^?{} ) break;                                                  $\C{// high priority}$
2720                or waitfor( girl )                                                              $\C{// girl called, compatible boy ? restart boy then girl}$
2721                        if ( ! is_empty( Boys[ccode] ) ) { `signal_block( Boys[ccode] );  signal_block( Girls[ccode] );` }
2722                or waitfor( boy ) {                                                             $\C{// boy called, compatible girl ? restart girl then boy}$
2723                        if ( ! is_empty( Girls[ccode] ) ) { `signal_block( Girls[ccode] );  signal_block( Boys[ccode] );` }
2724        }
2725}
2726\end{cfa}
2727\caption{Direct communication dating service}
2728\label{f:DirectCommunicationDatingService}
2729\end{figure}
2730
2731\begin{comment}
2732The following shows an example of two threads directly calling each other and accepting calls from each other in a cycle.
2733\begin{cfa}
2734\end{cfa}
2735\vspace{-0.8\baselineskip}
2736\begin{cquote}
2737\begin{tabular}{@{}l@{\hspace{3\parindentlnth}}l@{}}
2738\begin{cfa}
2739thread Ping {} pi;
2740void ping( Ping & mutex ) {}
2741void main( Ping & pi ) {
2742        for ( 10 ) {
2743                `waitfor( ping : pi );`
2744                `pong( po );`
2745        }
2746}
2747int main() {}
2748\end{cfa}
2749&
2750\begin{cfa}
2751thread Pong {} po;
2752void pong( Pong & mutex ) {}
2753void main( Pong & po ) {
2754        for ( 10 ) {
2755                `ping( pi );`
2756                `waitfor( pong : po );`
2757        }
2758}
2759
2760\end{cfa}
2761\end{tabular}
2762\end{cquote}
2763% \lstMakeShortInline@%
2764% \caption{Threads ping/pong using external scheduling}
2765% \label{f:pingpong}
2766% \end{figure}
2767Note, the ping/pong threads are globally declared, @pi@/@po@, and hence, start and possibly complete before the program main starts.
2768\end{comment}
2769
2770
2771\subsection{\texorpdfstring{\protect\lstinline@mutex@ Generators / coroutines / threads}{monitor Generators / coroutines / threads}}
2772
2773\CFA generators, coroutines, and threads can also be @mutex@ (Table~\ref{t:ExecutionPropertyComposition} cases 4, 6, 12) allowing safe \emph{direct communication} with threads, \ie the custom types can have mutex functions that are called by other threads.
2774All monitor features are available within these mutex functions.
2775For example, if the formatter generator or coroutine equivalent in Figure~\ref{f:CFAFormatGen} is extended with the monitor property and this interface function is used to communicate with the formatter:
2776\begin{cfa}
2777void fmt( Fmt & mutex fmt, char ch ) { fmt.ch = ch; resume( fmt ) }
2778\end{cfa}
2779multiple threads can safely pass characters for formatting.
2780
2781Figure~\ref{f:DirectCommunicationComparison} shows a comparison of direct call-communication in \CFA versus indirect channel-communication in Go.
2782(Ada has a similar mechanism to \CFA direct communication.)
2783% The thread main function is by default @mutex@, so the @mutex@ qualifier for the thread parameter is optional.
2784% The reason is that the thread logically starts instantaneously in the thread main acquiring its mutual exclusion, so it starts before any calls to prepare for synchronizing these calls.
2785The \CFA program @main@ uses the call/return paradigm to directly communicate with the @GoRtn main@, whereas Go switches to the unbuffered channel paradigm to indirectly communicate with the goroutine.
2786Communication by multiple threads is safe for the @gortn@ thread via mutex calls in \CFA or channel assignment in Go.
2787The difference between call and channel send occurs for buffered channels making the send asynchronous.
2788In \CFA, asynchronous call and multiple buffers are provided using an administrator and worker threads~\cite{Gentleman81} and/or futures (not discussed).
2789
2790Figure~\ref{f:DirectCommunicationDatingService} shows the dating-service problem in Figure~\ref{f:DatingServiceMonitor} extended from indirect monitor communication to direct thread communication.
2791When converting a monitor to a thread (server), the coding pattern is to move as much code as possible from the accepted functions into the thread main so it does as much work as possible.
2792Notice, the dating server is postponing requests for an unspecified time while continuing to accept new requests.
2793For complex servers, \eg web-servers, there can be hundreds of lines of code in the thread main and safe interaction with clients can be complex.
2794
2795
2796\subsection{Low-level Locks}
2797
2798For 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.
2799Some of these low-level mechanisms are used to build the \CFA runtime, but we always advocate using high-level mechanisms whenever possible.
2800
2801
2802% \section{Parallelism}
2803% \label{s:Parallelism}
2804%
2805% Historically, computer performance was about processor speeds.
2806% However, with heat dissipation being a direct consequence of speed increase, parallelism is the new source for increased performance~\cite{Sutter05, Sutter05b}.
2807% Therefore, high-performance applications must care about parallelism, which requires concurrency.
2808% The lowest-level approach of parallelism is to use \newterm{kernel threads} in combination with semantics like @fork@, @join@, \etc.
2809% However, kernel threads are better as an implementation tool because of complexity and higher cost.
2810% Therefore, different abstractions are often layered onto kernel threads to simplify them, \eg pthreads.
2811%
2812%
2813% \subsection{User threads}
2814%
2815% A direct improvement on kernel threads is user threads, \eg Erlang~\cite{Erlang} and \uC~\cite{uC++book}.
2816% 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.
2817% In many cases, user threads can be used on a much larger scale (100,000 threads).
2818% Like kernel threads, user threads support preemption, which maximizes nondeterminism, but increases the potential for concurrency errors: race, livelock, starvation, and deadlock.
2819% \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}.
2820%
2821% A variant of user thread is \newterm{fibres}, which removes preemption, \eg Go~\cite{Go} @goroutine@s.
2822% 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.
2823% However, preemption is necessary for fairness and to reduce tail-latency.
2824% For concurrency that relies on spinning, if all cores spin the system is livelocked, whereas preemption breaks the livelock.
2825
2826
2827\begin{comment}
2828\subsection{Thread pools}
2829
2830In 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.
2831If the jobs are dependent, \ie interact, there is an implicit dependency graph that ties them together.
2832While removing direct concurrency, and hence the amount of context switching, thread pools significantly limit the interaction that can occur among jobs.
2833Indeed, jobs should not block because that also blocks the underlying thread, which effectively means the CPU utilization, and therefore throughput, suffers.
2834While 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.
2835As well, concurrency errors return, which threads pools are suppose to mitigate.
2836
2837\begin{figure}
2838\centering
2839\begin{tabular}{@{}l|l@{}}
2840\begin{cfa}
2841struct Adder {
2842        int * row, cols;
2843};
2844int operator()() {
2845        subtotal = 0;
2846        for ( int c = 0; c < cols; c += 1 )
2847                subtotal += row[c];
2848        return subtotal;
2849}
2850void ?{}( Adder * adder, int row[$\,$], int cols, int & subtotal ) {
2851        adder.[rows, cols, subtotal] = [rows, cols, subtotal];
2852}
2853
2854
2855
2856
2857\end{cfa}
2858&
2859\begin{cfa}
2860int main() {
2861        const int rows = 10, cols = 10;
2862        int matrix[rows][cols], subtotals[rows], total = 0;
2863        // read matrix
2864        Executor executor( 4 ); // kernel threads
2865        Adder * adders[rows];
2866        for ( r; rows ) { // send off work for executor
2867                adders[r] = new( matrix[r], cols, &subtotal[r] );
2868                executor.send( *adders[r] );
2869        }
2870        for ( r; rows ) {       // wait for results
2871                delete( adders[r] );
2872                total += subtotals[r];
2873        }
2874        sout | total;
2875}
2876\end{cfa}
2877\end{tabular}
2878\caption{Executor}
2879\end{figure}
2880\end{comment}
2881
2882
2883\section{Runtime Structure}
2884\label{s:CFARuntimeStructure}
2885
2886Figure~\ref{f:RunTimeStructure} illustrates the runtime structure of a \CFA program.
2887In 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.
2888An executing thread is illustrated by its containment in a processor.
2889
2890\begin{figure}
2891\centering
2892\input{RunTimeStructure}
2893\caption{\CFA Runtime structure}
2894\label{f:RunTimeStructure}
2895\end{figure}
2896
2897
2898\subsection{Cluster}
2899\label{s:RuntimeStructureCluster}
2900
2901A \newterm{cluster} is a collection of user and kernel threads, where the kernel threads run the user threads from the cluster's ready queue, and the operating system runs the kernel threads on the processors from its ready queue~\cite{Buhr90a}.
2902The term \newterm{virtual processor} is introduced as a synonym for kernel thread to disambiguate between user and kernel thread.
2903From the language perspective, a virtual processor is an actual processor (core).
2904
2905The purpose of a cluster is to control the amount of parallelism that is possible among threads, plus scheduling and other execution defaults.
2906The default cluster-scheduler is single-queue multi-server, which provides automatic load-balancing of threads on processors.
2907However, the design allows changing the scheduler, \eg multi-queue multiserver with work-stealing/sharing across the virtual processors.
2908If several clusters exist, both threads and virtual processors, can be explicitly migrated from one cluster to another.
2909No automatic load balancing among clusters is performed by \CFA.
2910
2911When 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.
2912The user cluster is created to contain the application user-threads.
2913Having all threads execute on the one cluster often maximizes utilization of processors, which minimizes runtime.
2914However, because of limitations of scheduling requirements (real-time), NUMA architecture, heterogeneous hardware, or issues with the underlying operating system, multiple clusters are sometimes necessary.
2915
2916
2917\subsection{Virtual processor}
2918\label{s:RuntimeStructureProcessor}
2919
2920A 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.
2921Programs may use more virtual processors than hardware processors.
2922On a multiprocessor, kernel threads are distributed across the hardware processors resulting in virtual processors executing in parallel.
2923(It is possible to use affinity to lock a virtual processor onto a particular hardware processor~\cite{affinityLinux,affinityWindows}, which is used when caching issues occur or for heterogeneous hardware processors.) %, affinityFreebsd, affinityNetbsd, affinityMacosx
2924The \CFA runtime attempts to block unused processors and unblock processors as the system load increases;
2925balancing the workload with processors is difficult because it requires future knowledge, \ie what will the application workload do next.
2926Preemption occurs on virtual processors rather than user threads, via operating-system interrupts.
2927Thus virtual processors execute user threads, where preemption frequency applies to a virtual processor, so preemption occurs randomly across the executed user threads.
2928Turning off preemption transforms user threads into fibres.
2929
2930
2931\begin{comment}
2932\section{Implementation}
2933\label{s:Implementation}
2934
2935A primary implementation challenge is avoiding contention from dynamically allocating memory because of bulk acquire, \eg the internal-scheduling design is almost free of allocations.
2936All 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.
2937Furthermore, several bulk-acquire operations need a variable amount of memory.
2938This 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.
2939
2940In \CFA, ordering of monitor acquisition relies on memory ordering to prevent deadlock~\cite{Havender68}, because all objects have distinct nonoverlapping memory layouts, and mutual-exclusion for a monitor is only defined for its lifetime.
2941When a mutex call is made, pointers to the concerned monitors are aggregated into a variable-length array and sorted.
2942This array persists for the entire duration of the mutual exclusion and is used extensively for synchronization operations.
2943
2944To 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;
2945the corresponding registers are then restored for the other context.
2946Note, the instruction pointer is untouched since the context switch is always inside the same function.
2947Experimental 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.
2948
2949All kernel threads (@pthreads@) created a stack.
2950Each \CFA virtual processor is implemented as a coroutine and these coroutines run directly on the kernel-thread stack, effectively stealing this stack.
2951The exception to this rule is the program main, \ie the initial kernel thread that is given to any program.
2952In 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.
2953\end{comment}
2954
2955
2956\subsection{Preemption}
2957
2958Nondeterministic 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.
2959This atomic reliance can fail on multicore machines, because execution across cores is nondeterministic.
2960A different reason for not supporting preemption is that it significantly complicates the runtime system, \eg Windows runtime does not support interrupts and on Linux systems, interrupts are complex (see below).
2961Preemption is normally handled by setting a countdown timer on each virtual processor.
2962When the timer expires, an interrupt is delivered, and its signal 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.
2963Multiple signal handlers may be pending.
2964When 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.
2965The 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;
2966therefore, the same signal mask is required for all virtual processors in a cluster.
2967Because preemption interval is usually long (1 ms) performance cost is negligible.
2968
2969Linux switched a decade ago from specific to arbitrary virtual-processor signal-delivery for applications with multiple kernel threads.
2970In the new semantics, a virtual-processor directed signal may be delivered to any virtual processor created by the application that does not have the signal blocked.
2971Hence, 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).
2972To 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.
2973Virtual processors register an expiration time with the discrete-event simulator, which is inserted in sorted order.
2974The 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.
2975Processing a preemption event sends an \emph{internal} @SIGUSR1@ signal to the registered virtual processor, which is always delivered to that processor.
2976
2977
2978\subsection{Debug kernel}
2979
2980There are two versions of the \CFA runtime kernel: debug and nondebug.
2981The debugging version has many runtime checks and internal assertions, \eg stack nonwritable guard page, and checks for stack overflow whenever context switches occur among coroutines and threads, which catches most stack overflows.
2982After a program is debugged, the nondebugging version can be used to significantly decrease space and increase performance.
2983
2984
2985\section{Performance}
2986\label{s:Performance}
2987
2988To test the performance of the \CFA runtime, a series of microbenchmarks are used to compare \CFA with pthreads, Java 11.0.6, Go 1.12.6, Rust 1.37.0, Python 3.7.6, Node.js 12.14.1, and \uC 7.0.0.
2989For comparison, the package must be multiprocessor (M:N), which excludes libdil and libmil~\cite{libdill} (M:1)), and use a shared-memory programming model, \eg not message passing.
2990The benchmark computer is an AMD Opteron\texttrademark\ 6380 NUMA 64-core, 8 socket, 2.5 GHz processor, running Ubuntu 16.04.6 LTS, and pthreads/\CFA/\uC are compiled with gcc 9.2.1.
2991
2992All benchmarks are run using the following harness.
2993(The Java harness is augmented to circumvent JIT issues.)
2994\begin{cfa}
2995#define BENCH( `run` ) uint64_t start = cputime_ns();  `run;`  double result = (double)(cputime_ns() - start) / N;
2996\end{cfa}
2997where CPU time in nanoseconds is from the appropriate language clock.
2998Each benchmark is performed @N@ times, where @N@ is selected so the benchmark runs in the range of 2--20 s for the specific programming language;
2999each @N@ appears after the experiment name in the following tables.
3000The total time is divided by @N@ to obtain the average time for a benchmark.
3001Each benchmark experiment is run 13 times and the average appears in the table.
3002For languages with a runtime JIT (Java, Node.js, Python), a single half-hour long experiment is run to check stability;
3003all long-experiment results are statistically equivalent, \ie median/average/SD correlate with the short-experiment results, indicating the short experiments reached a steady state.
3004All omitted tests for other languages are functionally identical to the \CFA tests and available online~\cite{CforallConcurrentBenchmarks}.
3005
3006\subsection{Creation}
3007
3008Creation is measured by creating and deleting a specific kind of control-flow object.
3009Figure~\ref{f:creation} shows the code for \CFA with results in Table~\ref{t:creation}.
3010Note, the call stacks of \CFA coroutines are lazily created on the first resume, therefore the cost of creation with and without a stack are presented.
3011
3012\begin{multicols}{2}
3013\begin{cfa}[xleftmargin=0pt]
3014`coroutine` MyCoroutine {};
3015void ?{}( MyCoroutine & this ) {
3016#ifdef EAGER
3017        resume( this );
3018#endif
3019}
3020void main( MyCoroutine & ) {}
3021int main() {
3022        BENCH( for ( N ) { `MyCoroutine c;` } )
3023        sout | result;
3024}
3025\end{cfa}
3026\captionof{figure}{\CFA creation benchmark}
3027\label{f:creation}
3028
3029\columnbreak
3030
3031\vspace*{-16pt}
3032\captionof{table}{Creation comparison (nanoseconds)}
3033\label{t:creation}
3034
3035\begin{tabular}[t]{@{}r*{3}{D{.}{.}{5.2}}@{}}
3036\multicolumn{1}{@{}r}{Object(N)\hspace*{10pt}} & \multicolumn{1}{c}{Median} & \multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
3037\CFA generator (1B)                     & 0.6           & 0.6           & 0.0           \\
3038\CFA coroutine lazy     (100M)  & 13.4          & 13.1          & 0.5           \\
3039\CFA coroutine eager (10M)      & 144.7         & 143.9         & 1.5           \\
3040\CFA thread (10M)                       & 466.4         & 468.0         & 11.3          \\
3041\uC coroutine (10M)                     & 155.6         & 155.7         & 1.7           \\
3042\uC thread (10M)                        & 523.4         & 523.9         & 7.7           \\
3043Python generator (10M)          & 123.2         & 124.3         & 4.1           \\
3044Node.js generator (10M)         & 33.4          & 33.5          & 0.3           \\
3045Goroutine thread (10M)          & 751.0         & 750.5         & 3.1           \\
3046Rust tokio thread (10M)         & 1860.0        & 1881.1        & 37.6          \\
3047Rust thread     (250K)                  & 53801.0       & 53896.8       & 274.9         \\
3048Java thread (250K)                      & 119256.0      & 119679.2      & 2244.0        \\
3049% Java thread (1 000 000)               & 123100.0      & 123052.5      & 751.6         \\
3050Pthreads thread (250K)          & 31465.5       & 31419.5       & 140.4
3051\end{tabular}
3052\end{multicols}
3053
3054\vspace*{-10pt}
3055\subsection{Internal scheduling}
3056
3057Internal scheduling is measured using a cycle of two threads signaling and waiting.
3058Figure~\ref{f:schedint} shows the code for \CFA, with results in Table~\ref{t:schedint}.
3059Note, the \CFA incremental cost for bulk acquire is a fixed cost for small numbers of mutex objects.
3060User-level threading has one kernel thread, eliminating contention between the threads (direct handoff of the kernel thread).
3061Kernel-level threading has two kernel threads allowing some contention.
3062
3063\begin{multicols}{2}
3064\setlength{\tabcolsep}{3pt}
3065\begin{cfa}[xleftmargin=0pt]
3066volatile int go = 0;
3067`condition c;`
3068`monitor` M {} m1/*, m2, m3, m4*/;
3069void call( M & `mutex p1/*, p2, p3, p4*/` ) {
3070        `signal( c );`
3071}
3072void wait( M & `mutex p1/*, p2, p3, p4*/` ) {
3073        go = 1; // continue other thread
3074        for ( N ) { `wait( c );` } );
3075}
3076thread T {};
3077void main( T & ) {
3078        while ( go == 0 ) { yield(); } // waiter must start first
3079        BENCH( for ( N ) { call( m1/*, m2, m3, m4*/ ); } )
3080        sout | result;
3081}
3082int main() {
3083        T t;
3084        wait( m1/*, m2, m3, m4*/ );
3085}
3086\end{cfa}
3087\vspace*{-8pt}
3088\captionof{figure}{\CFA Internal-scheduling benchmark}
3089\label{f:schedint}
3090
3091\columnbreak
3092
3093\vspace*{-16pt}
3094\captionof{table}{Internal-scheduling comparison (nanoseconds)}
3095\label{t:schedint}
3096\bigskip
3097
3098\begin{tabular}{@{}r*{3}{D{.}{.}{5.2}}@{}}
3099\multicolumn{1}{@{}r}{Object(N)\hspace*{10pt}} & \multicolumn{1}{c}{Median} & \multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
3100\CFA @signal@, 1 monitor (10M)  & 364.4         & 364.2         & 4.4           \\
3101\CFA @signal@, 2 monitor (10M)  & 484.4         & 483.9         & 8.8           \\
3102\CFA @signal@, 4 monitor (10M)  & 709.1         & 707.7         & 15.0          \\
3103\uC @signal@ monitor (10M)              & 328.3         & 327.4         & 2.4           \\
3104Rust cond. variable     (1M)            & 7514.0        & 7437.4        & 397.2         \\
3105Java @notify@ monitor (1M)              & 8717.0        & 8774.1        & 471.8         \\
3106% Java @notify@ monitor (100 000 000)           & 8634.0        & 8683.5        & 330.5         \\
3107Pthreads cond. variable (1M)    & 5553.7        & 5576.1        & 345.6
3108\end{tabular}
3109\end{multicols}
3110
3111
3112\subsection{External scheduling}
3113
3114External scheduling is measured using a cycle of two threads calling and accepting the call using the @waitfor@ statement.
3115Figure~\ref{f:schedext} shows the code for \CFA with results in Table~\ref{t:schedext}.
3116Note, the \CFA incremental cost for bulk acquire is a fixed cost for small numbers of mutex objects.
3117
3118\begin{multicols}{2}
3119\setlength{\tabcolsep}{5pt}
3120\vspace*{-16pt}
3121\begin{cfa}[xleftmargin=0pt]
3122`monitor` M {} m1/*, m2, m3, m4*/;
3123void call( M & `mutex p1/*, p2, p3, p4*/` ) {}
3124void wait( M & `mutex p1/*, p2, p3, p4*/` ) {
3125        for ( N ) { `waitfor( call : p1/*, p2, p3, p4*/ );` }
3126}
3127thread T {};
3128void main( T & ) {
3129        BENCH( for ( N ) { call( m1/*, m2, m3, m4*/ ); } )
3130        sout | result;
3131}
3132int main() {
3133        T t;
3134        wait( m1/*, m2, m3, m4*/ );
3135}
3136\end{cfa}
3137\captionof{figure}{\CFA external-scheduling benchmark}
3138\label{f:schedext}
3139
3140\columnbreak
3141
3142\vspace*{-18pt}
3143\captionof{table}{External-scheduling comparison (nanoseconds)}
3144\label{t:schedext}
3145\begin{tabular}{@{}r*{3}{D{.}{.}{3.2}}@{}}
3146\multicolumn{1}{@{}r}{Object(N)\hspace*{10pt}} & \multicolumn{1}{c}{Median} &\multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
3147\CFA @waitfor@, 1 monitor (10M) & 367.1 & 365.3 & 5.0   \\
3148\CFA @waitfor@, 2 monitor (10M) & 463.0 & 464.6 & 7.1   \\
3149\CFA @waitfor@, 4 monitor (10M) & 689.6 & 696.2 & 21.5  \\
3150\uC \lstinline[language=uC++]|_Accept| monitor (10M)    & 328.2 & 329.1 & 3.4   \\
3151Go \lstinline[language=Golang]|select| channel (10M)    & 365.0 & 365.5 & 1.2
3152\end{tabular}
3153\end{multicols}
3154
3155\subsection{Mutual-Exclusion}
3156
3157Uncontented mutual exclusion, which frequently occurs, is measured by entering and leaving a critical section.
3158For monitors, entering and leaving a mutex function are measured, otherwise the language-appropriate mutex-lock is measured.
3159For comparison, a spinning (vs.\ blocking) test-and-test-set lock is presented.
3160Figure~\ref{f:mutex} shows the code for \CFA with results in Table~\ref{t:mutex}.
3161Note the incremental cost of bulk acquire for \CFA, which is largely a fixed cost for small numbers of mutex objects.
3162
3163\begin{multicols}{2}
3164\setlength{\tabcolsep}{3pt}
3165\begin{cfa}[xleftmargin=0pt]
3166`monitor` M {} m1/*, m2, m3, m4*/;
3167call( M & `mutex p1/*, p2, p3, p4*/` ) {}
3168int main() {
3169        BENCH( for( N ) call( m1/*, m2, m3, m4*/ ); )
3170        sout | result;
3171}
3172\end{cfa}
3173\captionof{figure}{\CFA acquire/release mutex benchmark}
3174\label{f:mutex}
3175
3176\columnbreak
3177
3178\vspace*{-16pt}
3179\captionof{table}{Mutex comparison (nanoseconds)}
3180\label{t:mutex}
3181\begin{tabular}{@{}r*{3}{D{.}{.}{3.2}}@{}}
3182\multicolumn{1}{@{}r}{Object(N)\hspace*{10pt}} & \multicolumn{1}{c}{Median} &\multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
3183test-and-test-set lock (50M)            & 19.1  & 18.9  & 0.4   \\
3184\CFA @mutex@ function, 1 arg. (50M)     & 48.3  & 47.8  & 0.9   \\
3185\CFA @mutex@ function, 2 arg. (50M)     & 86.7  & 87.6  & 1.9   \\
3186\CFA @mutex@ function, 4 arg. (50M)     & 173.4 & 169.4 & 5.9   \\
3187\uC @monitor@ member rtn. (50M)         & 54.8  & 54.8  & 0.1   \\
3188Goroutine mutex lock (50M)                      & 34.0  & 34.0  & 0.0   \\
3189Rust mutex lock (50M)                           & 33.0  & 33.2  & 0.8   \\
3190Java synchronized method (50M)          & 31.0  & 30.9  & 0.5   \\
3191% Java synchronized method (10 000 000 000)             & 31.0 & 30.2 & 0.9 \\
3192Pthreads mutex Lock (50M)                       & 31.0  & 31.1  & 0.4
3193\end{tabular}
3194\end{multicols}
3195
3196\subsection{Context switching}
3197
3198In procedural programming, the cost of a function call is important as modularization (refactoring) increases.
3199(In many cases, a compiler inlines function calls to increase the size and number of basic blocks for optimizing.)
3200Similarly, when modularization extends to coroutines and threads, the time for a context switch becomes a relevant factor.
3201The coroutine test is from resumer to suspender and from suspender to resumer, which is two context switches.
3202%For async-await systems, the test is scheduling and fulfilling @N@ empty promises, where all promises are allocated before versus interleaved with fulfillment to avoid garbage collection.
3203For async-await systems, the test measures the cost of the @await@ expression entering the event engine by awaiting @N@ promises, where each created promise is resolved by an immediate event in the engine (using Node.js @setImmediate@).
3204The thread test is using yield to enter and return from the runtime kernel, which is two context switches.
3205The difference in performance between coroutine and thread context-switch is the cost of scheduling for threads, whereas coroutines are self-scheduling.
3206Figure~\ref{f:ctx-switch} shows the \CFA code for a coroutine and thread with results in Table~\ref{t:ctx-switch}.
3207
3208% From: Gregor Richards <gregor.richards@uwaterloo.ca>
3209% To: "Peter A. Buhr" <pabuhr@plg2.cs.uwaterloo.ca>
3210% Date: Fri, 24 Jan 2020 13:49:18 -0500
3211%
3212% I can also verify that the previous version, which just tied a bunch of promises together, *does not* go back to the
3213% event loop at all in the current version of Node. Presumably they're taking advantage of the fact that the ordering of
3214% events is intentionally undefined to just jump right to the next 'then' in the chain, bypassing event queueing
3215% entirely. That's perfectly correct behavior insofar as its difference from the specified behavior isn't observable, but
3216% it isn't typical or representative of much anything useful, because most programs wouldn't have whole chains of eager
3217% promises. Also, it's not representative of *anything* you can do with async/await, as there's no way to encode such an
3218% eager chain that way.
3219
3220\begin{multicols}{2}
3221\begin{cfa}[xleftmargin=0pt]
3222`coroutine` C {};
3223void main( C & ) { for () { `suspend;` } }
3224int main() { // coroutine test
3225        C c;
3226        BENCH( for ( N ) { `resume( c );` } )
3227        sout | result;
3228}
3229int main() { // thread test
3230        BENCH( for ( N ) { `yield();` } )
3231        sout | result;
3232}
3233\end{cfa}
3234\captionof{figure}{\CFA context-switch benchmark}
3235\label{f:ctx-switch}
3236
3237\columnbreak
3238
3239\vspace*{-16pt}
3240\captionof{table}{Context switch comparison (nanoseconds)}
3241\label{t:ctx-switch}
3242\begin{tabular}{@{}r*{3}{D{.}{.}{3.2}}@{}}
3243\multicolumn{1}{@{}r}{Object(N)\hspace*{10pt}} & \multicolumn{1}{c}{Median} &\multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
3244C function (10B)                        & 1.8           & 1.8           & 0.0   \\
3245\CFA generator (5B)                     & 1.8           & 2.0           & 0.3   \\
3246\CFA coroutine (100M)           & 32.5          & 32.9          & 0.8   \\
3247\CFA thread (100M)                      & 93.8          & 93.6          & 2.2   \\
3248\uC coroutine (100M)            & 50.3          & 50.3          & 0.2   \\
3249\uC thread (100M)                       & 97.3          & 97.4          & 1.0   \\
3250Python generator (100M)         & 40.9          & 41.3          & 1.5   \\
3251Node.js await (5M)                      & 1852.2        & 1854.7        & 16.4  \\
3252Node.js generator (100M)        & 33.3          & 33.4          & 0.3   \\
3253Goroutine thread (100M)         & 143.0         & 143.3         & 1.1   \\
3254Rust async await (100M)         & 32.0          & 32.0          & 0.0   \\
3255Rust tokio thread (100M)        & 143.0         & 143.0         & 1.7   \\
3256Rust thread (25M)                       & 332.0         & 331.4         & 2.4   \\
3257Java thread (100M)                      & 405.0         & 415.0         & 17.6  \\
3258% Java thread (  100 000 000)                   & 413.0 & 414.2 & 6.2 \\
3259% Java thread (5 000 000 000)                   & 415.0 & 415.2 & 6.1 \\
3260Pthreads thread (25M)           & 334.3         & 335.2         & 3.9
3261\end{tabular}
3262\end{multicols}
3263
3264
3265\subsection{Discussion}
3266
3267Languages using 1:1 threading based on pthreads can at best meet or exceed, due to language overhead, the pthread results.
3268Note, pthreads has a fast zero-contention mutex lock checked in user space.
3269Languages with M:N threading have better performance than 1:1 because there is no operating-system interactions (context-switching or locking).
3270As well, for locking experiments, M:N threading has less contention if only one kernel thread is used.
3271Languages with stackful coroutines have higher cost than stackless coroutines because of stack allocation and context switching;
3272however, stackful \uC and \CFA coroutines have approximately the same performance as stackless Python and Node.js generators.
3273The \CFA stackless generator is approximately 25 times faster for suspend/resume and 200 times faster for creation than stackless Python and Node.js generators.
3274The Node.js context-switch is costly when asynchronous await must enter the event engine because a promise is not fulfilled.
3275Finally, the benchmark results correlate across programming languages with and without JIT, indicating the JIT has completed any runtime optimizations.
3276
3277
3278\section{Conclusions and Future Work}
3279
3280Advanced control-flow will always be difficult, especially when there is temporal ordering and nondeterminism.
3281However, many systems exacerbate the difficulty through their presentation mechanisms.
3282This paper shows it is possible to understand high-level control-flow using three properties: statefulness, thread, mutual-exclusion/synchronization.
3283Combining these properties creates a number of high-level, efficient, and maintainable control-flow types: generator, coroutine, thread, each of which can be a monitor.
3284Eliminated from \CFA are barging and spurious wakeup, which are nonintuitive and lead to errors, and having to work with a bewildering set of low-level locks and acquisition techniques.
3285\CFA high-level race-free monitors and threads, when used with mutex access function, provide the core mechanisms for mutual exclusion and synchronization, without having to resort to magic qualifiers like @volatile@ or @atomic@.
3286Extending these mechanisms to handle high-level deadlock-free bulk acquire across both mutual exclusion and synchronization is a unique contribution.
3287The \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.
3288The M:N model is judged to be efficient and provide greater flexibility than a 1:1 threading model.
3289These 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.
3290Performance comparisons with other concurrent systems and languages show the \CFA approach is competitive across all basic operations, which translates directly into good performance in well-written applications with advanced control-flow.
3291C 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 using only calling communication.
3292
3293While control flow in \CFA has a strong start, development is still underway to complete a number of missing features.
3294
3295\medskip
3296\textbf{Flexible scheduling:}
3297An important part of concurrency is scheduling.
3298Different scheduling algorithms can affect performance, both in terms of average and variation.
3299However, no single scheduler is optimal for all workloads and therefore there is value in being able to change the scheduler for given programs.
3300One solution is to offer various tuning options, allowing the scheduler to be adjusted to the requirements of the workload.
3301However, to be truly flexible, a pluggable scheduler is necessary.
3302Currently, 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}.
3303
3304\smallskip
3305\textbf{Non-Blocking I/O:}
3306Many modern workloads are not bound by computation but IO operations, common cases being web servers and XaaS~\cite{XaaS} (anything as a service).
3307These types of workloads require significant engineering to amortizing costs of blocking IO-operations.
3308At its core, nonblocking 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.
3309Current 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.
3310However, these solutions lead to code that is hard to create, read, and maintain.
3311A better approach is to tie nonblocking I/O into the concurrency system to provide ease of use with low overhead, \eg thread-per-connection web-services.
3312A nonblocking I/O library is currently under development for \CFA.
3313
3314\smallskip
3315\textbf{Other concurrency tools:}
3316While monitors offer flexible and powerful concurrency for \CFA, other concurrency tools are also necessary for a complete multi-paradigm concurrency package.
3317Examples of such tools can include futures and promises~\cite{promises}, executors and actors.
3318These additional features are useful for applications that can be constructed without shared data and direct blocking.
3319As well, new \CFA extensions should make it possible to create a uniform interface for virtually all mutual exclusion, including monitors and low-level locks.
3320
3321\smallskip
3322\textbf{Implicit threading:}
3323Basic \emph{embarrassingly parallel} applications can benefit greatly from implicit concurrency, where sequential programs are converted to concurrent, with some help from pragmas to guide the conversion.
3324This type of concurrency can be achieved both at the language level and at the library level.
3325The canonical example of implicit concurrency is concurrent nested @for@ loops, which are amenable to divide and conquer algorithms~\cite{uC++book}.
3326The \CFA language features should make it possible to develop a reasonable number of implicit concurrency mechanisms to solve basic HPC data-concurrency problems.
3327However, implicit concurrency is a restrictive solution with significant limitations, so it can never replace explicit concurrent programming.
3328
3329
3330\section{Acknowledgements}
3331
3332The authors recognize the design assistance of Aaron Moss, Rob Schluntz, Andrew Beach, and Michael Brooks; David Dice for commenting and helping with the Java benchmarks; and Gregor Richards for helping with the Node.js benchmarks.
3333This research is funded by the NSERC/Waterloo-Huawei (\url{http://www.huawei.com}) Joint Innovation Lab. %, and Peter Buhr is partially funded by the Natural Sciences and Engineering Research Council of Canada.
3334
3335{%
3336\fontsize{9bp}{11.5bp}\selectfont%
3337\bibliography{pl,local}
3338}%
3339
3340\end{document}
3341
3342% Local Variables: %
3343% tab-width: 4 %
3344% fill-column: 120 %
3345% compile-command: "make" %
3346% End: %
Note: See TracBrowser for help on using the repository browser.