source: doc/papers/concurrency/Paper.tex @ b4d34fa

ADTarm-ehast-experimentalenumforall-pointer-decayjacob/cs343-translationjenkins-sandboxnew-astnew-ast-unique-exprpthread-emulationqualifiedEnum
Last change on this file since b4d34fa was d7a02ae, checked in by Peter A. Buhr <pabuhr@…>, 5 years ago

first complete draft of new concurrency paper

  • Property mode set to 100644
File size: 152.8 KB
Line 
1\documentclass[AMA,STIX1COL]{WileyNJD-v2}
2
3\articletype{RESEARCH ARTICLE}%
4
5\received{XXXXX}
6\revised{XXXXX}
7\accepted{XXXXX}
8
9\raggedbottom
10
11%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
12
13% Latex packages used in the document.
14
15\usepackage{epic,eepic}
16\usepackage{xspace}
17\usepackage{comment}
18\usepackage{upquote}                                            % switch curled `'" to straight
19\usepackage{listings}                                           % format program code
20\usepackage[labelformat=simple,aboveskip=0pt,farskip=0pt]{subfig}
21\renewcommand{\thesubfigure}{(\Alph{subfigure})}
22\captionsetup{justification=raggedright,singlelinecheck=false}
23\usepackage{dcolumn}                                            % align decimal points in tables
24\usepackage{capt-of}
25
26\hypersetup{breaklinks=true}
27\definecolor{OliveGreen}{cmyk}{0.64 0 0.95 0.40}
28\definecolor{Mahogany}{cmyk}{0 0.85 0.87 0.35}
29\definecolor{Plum}{cmyk}{0.50 1 0 0}
30
31\usepackage[pagewise]{lineno}
32\renewcommand{\linenumberfont}{\scriptsize\sffamily}
33
34\renewcommand{\topfraction}{0.8}                        % float must be greater than X of the page before it is forced onto its own page
35\renewcommand{\bottomfraction}{0.8}                     % float must be greater than X of the page before it is forced onto its own page
36\renewcommand{\floatpagefraction}{0.8}          % float must be greater than X of the page before it is forced onto its own page
37\renewcommand{\textfraction}{0.0}                       % the entire page maybe devoted to floats with no text on the page at all
38
39\lefthyphenmin=3                                                        % hyphen only after 4 characters
40\righthyphenmin=3
41
42%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
43
44% Names used in the document.
45
46\newcommand{\CFAIcon}{\textsf{C}\raisebox{\depth}{\rotatebox{180}{\textsf{A}}}\xspace} % Cforall symbolic name
47\newcommand{\CFA}{\protect\CFAIcon}             % safe for section/caption
48\newcommand{\CFL}{\textrm{Cforall}\xspace}      % Cforall symbolic name
49\newcommand{\Celeven}{\textrm{C11}\xspace}      % C11 symbolic name
50\newcommand{\CC}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}\xspace} % C++ symbolic name
51\newcommand{\CCeleven}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}11\xspace} % C++11 symbolic name
52\newcommand{\CCfourteen}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}14\xspace} % C++14 symbolic name
53\newcommand{\CCseventeen}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}17\xspace} % C++17 symbolic name
54\newcommand{\CCtwenty}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}20\xspace} % C++20 symbolic name
55\newcommand{\Csharp}{C\raisebox{-0.7ex}{\Large$^\sharp$}\xspace} % C# symbolic name
56
57%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
58
59\newcommand{\Textbf}[2][red]{{\color{#1}{\textbf{#2}}}}
60\newcommand{\Emph}[2][red]{{\color{#1}\textbf{\emph{#2}}}}
61\newcommand{\uC}{$\mu$\CC}
62\newcommand{\TODO}[1]{{\Textbf{#1}}}
63
64%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
65
66% Default underscore is too low and wide. Cannot use lstlisting "literate" as replacing underscore
67% removes it as a variable-name character so keywords in variables are highlighted. MUST APPEAR
68% AFTER HYPERREF.
69%\DeclareTextCommandDefault{\textunderscore}{\leavevmode\makebox[1.2ex][c]{\rule{1ex}{0.1ex}}}
70\renewcommand{\textunderscore}{\leavevmode\makebox[1.2ex][c]{\rule{1ex}{0.075ex}}}
71
72\renewcommand*{\thefootnote}{\Alph{footnote}} % hack because fnsymbol does not work
73%\renewcommand*{\thefootnote}{\fnsymbol{footnote}}
74
75\makeatletter
76% parindent is relative, i.e., toggled on/off in environments like itemize, so store the value for
77% use rather than use \parident directly.
78\newlength{\parindentlnth}
79\setlength{\parindentlnth}{\parindent}
80
81\newcommand{\LstBasicStyle}[1]{{\lst@basicstyle{\lst@basicstyle{#1}}}}
82\newcommand{\LstKeywordStyle}[1]{{\lst@basicstyle{\lst@keywordstyle{#1}}}}
83\newcommand{\LstCommentStyle}[1]{{\lst@basicstyle{\lst@commentstyle{#1}}}}
84
85\newlength{\gcolumnposn}                                        % temporary hack because lstlisting does not handle tabs correctly
86\newlength{\columnposn}
87\setlength{\gcolumnposn}{3.5in}
88\setlength{\columnposn}{\gcolumnposn}
89
90\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}}}}
91\newcommand{\CRT}{\global\columnposn=\gcolumnposn}
92
93% Denote newterms in particular font and index them without particular font and in lowercase, e.g., \newterm{abc}.
94% The option parameter provides an index term different from the new term, e.g., \newterm[\texttt{abc}]{abc}
95% The star version does not lowercase the index information, e.g., \newterm*{IBM}.
96\newcommand{\newtermFontInline}{\emph}
97\newcommand{\newterm}{\@ifstar\@snewterm\@newterm}
98\newcommand{\@newterm}[2][\@empty]{\lowercase{\def\temp{#2}}{\newtermFontInline{#2}}\ifx#1\@empty\index{\temp}\else\index{#1@{\protect#2}}\fi}
99\newcommand{\@snewterm}[2][\@empty]{{\newtermFontInline{#2}}\ifx#1\@empty\index{#2}\else\index{#1@{\protect#2}}\fi}
100
101% Latin abbreviation
102\newcommand{\abbrevFont}{\textit}                       % set empty for no italics
103\@ifundefined{eg}{
104\newcommand{\EG}{\abbrevFont{e}\abbrevFont{g}}
105\newcommand*{\eg}{%
106        \@ifnextchar{,}{\EG}%
107                {\@ifnextchar{:}{\EG}%
108                        {\EG,\xspace}}%
109}}{}%
110\@ifundefined{ie}{
111\newcommand{\IE}{\abbrevFont{i}\abbrevFont{e}}
112\newcommand*{\ie}{%
113        \@ifnextchar{,}{\IE}%
114                {\@ifnextchar{:}{\IE}%
115                        {\IE,\xspace}}%
116}}{}%
117\@ifundefined{etc}{
118\newcommand{\ETC}{\abbrevFont{etc}}
119\newcommand*{\etc}{%
120        \@ifnextchar{.}{\ETC}%
121        {\ETC.\xspace}%
122}}{}%
123\@ifundefined{etal}{
124\newcommand{\ETAL}{\abbrevFont{et}~\abbrevFont{al}}
125\newcommand*{\etal}{%
126        \@ifnextchar{.}{\protect\ETAL}%
127                {\protect\ETAL.\xspace}%
128}}{}%
129\@ifundefined{viz}{
130\newcommand{\VIZ}{\abbrevFont{viz}}
131\newcommand*{\viz}{%
132        \@ifnextchar{.}{\VIZ}%
133                {\VIZ.\xspace}%
134}}{}%
135\makeatother
136
137\newenvironment{cquote}
138               {\list{}{\lstset{resetmargins=true,aboveskip=0pt,belowskip=0pt}\topsep=3pt\parsep=0pt\leftmargin=\parindentlnth\rightmargin\leftmargin}%
139                \item\relax}
140               {\endlist}
141
142%\newenvironment{cquote}{%
143%\list{}{\lstset{resetmargins=true,aboveskip=0pt,belowskip=0pt}\topsep=3pt\parsep=0pt\leftmargin=\parindentlnth\rightmargin\leftmargin}%
144%\item\relax%
145%}{%
146%\endlist%
147%}% cquote
148
149% CFA programming language, based on ANSI C (with some gcc additions)
150\lstdefinelanguage{CFA}[ANSI]{C}{
151        morekeywords={
152                _Alignas, _Alignof, __alignof, __alignof__, asm, __asm, __asm__, __attribute, __attribute__,
153                auto, _Bool, catch, catchResume, choose, _Complex, __complex, __complex__, __const, __const__,
154                coroutine, disable, dtype, enable, exception, __extension__, fallthrough, fallthru, finally,
155                __float80, float80, __float128, float128, forall, ftype, generator, _Generic, _Imaginary, __imag, __imag__,
156                inline, __inline, __inline__, __int128, int128, __label__, monitor, mutex, _Noreturn, one_t, or,
157                otype, restrict, __restrict, __restrict__, __signed, __signed__, _Static_assert, suspend, thread,
158                _Thread_local, throw, throwResume, timeout, trait, try, ttype, typeof, __typeof, __typeof__,
159                virtual, __volatile, __volatile__, waitfor, when, with, zero_t},
160        moredirectives={defined,include_next},
161        % replace/adjust listing characters that look bad in sanserif
162        literate={-}{\makebox[1ex][c]{\raisebox{0.4ex}{\rule{0.8ex}{0.1ex}}}}1 {^}{\raisebox{0.6ex}{$\scriptstyle\land\,$}}1
163                {~}{\raisebox{0.3ex}{$\scriptstyle\sim\,$}}1 % {`}{\ttfamily\upshape\hspace*{-0.1ex}`}1
164                {<}{\textrm{\textless}}1 {>}{\textrm{\textgreater}}1
165                {<-}{$\leftarrow$}2 {=>}{$\Rightarrow$}2 {->}{\makebox[1ex][c]{\raisebox{0.5ex}{\rule{0.8ex}{0.075ex}}}\kern-0.2ex{\textrm{\textgreater}}}2,
166}
167
168\lstset{
169language=CFA,
170columns=fullflexible,
171basicstyle=\linespread{0.9}\sf,                                                 % reduce line spacing and use sanserif font
172stringstyle=\tt,                                                                                % use typewriter font
173tabsize=5,                                                                                              % N space tabbing
174xleftmargin=\parindentlnth,                                                             % indent code to paragraph indentation
175%mathescape=true,                                                                               % LaTeX math escape in CFA code $...$
176escapechar=\$,                                                                                  % LaTeX escape in CFA code
177keepspaces=true,                                                                                %
178showstringspaces=false,                                                                 % do not show spaces with cup
179showlines=true,                                                                                 % show blank lines at end of code
180aboveskip=4pt,                                                                                  % spacing above/below code block
181belowskip=3pt,
182moredelim=**[is][\color{red}]{`}{`},
183}% lstset
184
185% uC++ programming language, based on ANSI C++
186\lstdefinelanguage{uC++}[ANSI]{C++}{
187        morekeywords={
188                _Accept, _AcceptReturn, _AcceptWait, _Actor, _At, _CatchResume, _Cormonitor, _Coroutine, _Disable,
189                _Else, _Enable, _Event, _Finally, _Monitor, _Mutex, _Nomutex, _PeriodicTask, _RealTimeTask,
190                _Resume, _Select, _SporadicTask, _Task, _Timeout, _When, _With, _Throw},
191}
192\lstdefinelanguage{Golang}{
193        morekeywords=[1]{package,import,func,type,struct,return,defer,panic,recover,select,var,const,iota,},
194        morekeywords=[2]{string,uint,uint8,uint16,uint32,uint64,int,int8,int16,int32,int64,
195                bool,float32,float64,complex64,complex128,byte,rune,uintptr, error,interface},
196        morekeywords=[3]{map,slice,make,new,nil,len,cap,copy,close,true,false,delete,append,real,imag,complex,chan,},
197        morekeywords=[4]{for,break,continue,range,goto,switch,case,fallthrough,if,else,default,},
198        morekeywords=[5]{Println,Printf,Error,},
199        sensitive=true,
200        morecomment=[l]{//},
201        morecomment=[s]{/*}{*/},
202        morestring=[b]',
203        morestring=[b]",
204        morestring=[s]{`}{`},
205}
206
207% Go programming language: https://github.com/julienc91/listings-golang/blob/master/listings-golang.sty
208\lstdefinelanguage{Golang}{
209        morekeywords=[1]{package,import,func,type,struct,return,defer,panic,recover,select,var,const,iota,},
210        morekeywords=[2]{string,uint,uint8,uint16,uint32,uint64,int,int8,int16,int32,int64,
211                bool,float32,float64,complex64,complex128,byte,rune,uintptr, error,interface},
212        morekeywords=[3]{map,slice,make,new,nil,len,cap,copy,close,true,false,delete,append,real,imag,complex,chan,},
213        morekeywords=[4]{for,break,continue,range,goto,switch,case,fallthrough,if,else,default,},
214        morekeywords=[5]{Println,Printf,Error,},
215        sensitive=true,
216        morecomment=[l]{//},
217        morecomment=[s]{/*}{*/},
218        morestring=[b]',
219        morestring=[b]",
220        morestring=[s]{`}{`},
221        % replace/adjust listing characters that look bad in sanserif
222        literate={-}{\makebox[1ex][c]{\raisebox{0.4ex}{\rule{0.8ex}{0.1ex}}}}1 {^}{\raisebox{0.6ex}{$\scriptstyle\land\,$}}1
223                {~}{\raisebox{0.3ex}{$\scriptstyle\sim\,$}}1 % {`}{\ttfamily\upshape\hspace*{-0.1ex}`}1
224                {<}{\textrm{\textless}}1 {>}{\textrm{\textgreater}}1
225                {<-}{\makebox[2ex][c]{\textrm{\textless}\raisebox{0.5ex}{\rule{0.8ex}{0.075ex}}}}2,
226}
227
228\lstnewenvironment{cfa}[1][]
229{\lstset{#1}}
230{}
231\lstnewenvironment{C++}[1][]                            % use C++ style
232{\lstset{language=C++,moredelim=**[is][\protect\color{red}]{`}{`},#1}\lstset{#1}}
233{}
234\lstnewenvironment{uC++}[1][]
235{\lstset{#1}}
236{}
237\lstnewenvironment{Go}[1][]
238{\lstset{language=Golang,moredelim=**[is][\protect\color{red}]{`}{`},#1}\lstset{#1}}
239{}
240\lstnewenvironment{python}[1][]
241{\lstset{language=python,moredelim=**[is][\protect\color{red}]{`}{`},#1}\lstset{#1}}
242{}
243
244% inline code @...@
245\lstMakeShortInline@%
246
247\let\OLDthebibliography\thebibliography
248\renewcommand\thebibliography[1]{
249  \OLDthebibliography{#1}
250  \setlength{\parskip}{0pt}
251  \setlength{\itemsep}{4pt plus 0.3ex}
252}
253
254\newbox\myboxA
255\newbox\myboxB
256\newbox\myboxC
257\newbox\myboxD
258
259\title{\texorpdfstring{Advanced Control-flow and Concurrency in \protect\CFA}{Advanced Control-flow in Cforall}}
260
261\author[1]{Thierry Delisle}
262\author[1]{Peter A. Buhr*}
263\authormark{DELISLE \textsc{et al.}}
264
265\address[1]{\orgdiv{Cheriton School of Computer Science}, \orgname{University of Waterloo}, \orgaddress{\state{Waterloo, ON}, \country{Canada}}}
266
267\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}}
268
269% \fundingInfo{Natural Sciences and Engineering Research Council of Canada}
270
271\abstract[Summary]{
272\CFA is a polymorphic, non-object-oriented, concurrent, backwards-compatible extension of the C programming language.
273This paper discusses the design philosophy and implementation of its advanced control-flow and concurrent/parallel features, along with the supporting runtime.
274These features are created from scratch as ISO C has only low-level and/or unimplemented concurrency, so C programmers continue to rely on library features like C pthreads.
275\CFA introduces modern language-level control-flow mechanisms, like generators, coroutines, user-level threading, and monitors for mutual exclusion and synchronization.
276% Library extension for executors, futures, and actors are built on these basic mechanisms.
277The runtime provides significant programmer simplification and safety by eliminating spurious wakeup and optional monitor barging.
278The runtime also ensures multiple monitors can be safely acquired \emph{simultaneously} (deadlock free), and this feature is fully integrated with all monitor synchronization mechanisms.
279All control-flow features integrate with the \CFA polymorphic type-system and exception handling, while respecting the expectations and style of C programmers.
280Experimental results show comparable performance of the new features with similar mechanisms in other concurrent programming-languages.
281}%
282
283\keywords{generator, coroutine, concurrency, parallelism, thread, monitor, runtime, C, \CFA (Cforall)}
284
285
286\begin{document}
287\linenumbers                                            % comment out to turn off line numbering
288
289\maketitle
290
291
292\section{Introduction}
293
294This paper discusses the design philosophy and implementation of advanced language-level control-flow and concurrent/parallel features in \CFA~\cite{Moss18} and its runtime.
295\CFA is a modern, polymorphic, non-object-oriented\footnote{
296\CFA has features often associated with object-oriented programming languages, such as constructors, destructors, virtuals and simple inheritance.
297However, functions \emph{cannot} be nested in structures, so there is no lexical binding between a structure and set of functions (member/method) implemented by an implicit \lstinline@this@ (receiver) parameter.},
298backwards-compatible extension of the C programming language.
299In many ways, \CFA is to C as Scala~\cite{Scala} is to Java, providing a \emph{research vehicle} for new typing and control flow capabilities on top of a highly popular programming language allowing immediate dissemination.
300Within the \CFA framework, new control-flow features are created from scratch.
301ISO \Celeven defines only a subset of the \CFA extensions, where the overlapping features are concurrency~\cite[\S~7.26]{C11}.
302However, \Celeven concurrency is largely wrappers for a subset of the pthreads library~\cite{Butenhof97,Pthreads}.
303Furthermore, \Celeven and pthreads concurrency is simple, based on thread fork/join in a function and a few locks, which is low-level and error prone;
304no high-level language concurrency features are defined.
305Interestingly, almost a decade after publication of the \Celeven standard, neither gcc-8, clang-8 nor msvc-19 (most recent versions) support the \Celeven include @threads.h@, indicating little interest in the C11 concurrency approach.
306Finally, while the \Celeven standard does not state a threading model, the historical association with pthreads suggests implementations would adopt kernel-level threading (1:1)~\cite{ThreadModel}.
307
308In contrast, there has been a renewed interest during the past decade in user-level (M:N, green) threading in old and new programming languages.
309As multi-core hardware became available in the 1980/90s, both user and kernel threading were examined.
310Kernel threading was chosen, largely because of its simplicity and fit with the simpler operating systems and hardware architectures at the time, which gave it a performance advantage~\cite{Drepper03}.
311Libraries like pthreads were developed for C, and the Solaris operating-system switched from user (JDK 1.1~\cite{JDK1.1}) to kernel threads.
312As a result, languages like Java, Scala, Objective-C~\cite{obj-c-book}, \CCeleven~\cite{C11}, and C\#~\cite{Csharp} adopt the 1:1 kernel-threading model, with a variety of presentation mechanisms.
313From 2000 onwards, languages like Go~\cite{Go}, Erlang~\cite{Erlang}, Haskell~\cite{Haskell}, D~\cite{D}, and \uC~\cite{uC++,uC++book} have championed the M:N user-threading model, and many user-threading libraries have appeared~\cite{Qthreads,MPC,BoostThreads}, including putting green threads back into Java~\cite{Quasar}.
314The main argument for user-level threading is that they are lighter weight than kernel threads (locking and context switching do not cross the kernel boundary), so there is less restriction on programming styles that encourage large numbers of threads performing medium work-units to facilitate load balancing by the runtime~\cite{Verch12}.
315As well, user-threading facilitates a simpler concurrency approach using thread objects that leverage sequential patterns versus events with call-backs~\cite{vonBehren03}.
316Finally, performant user-threading implementations (both time and space) are largely competitive with direct kernel-threading implementations, while achieving the programming advantages of high concurrency levels and safety.
317
318A further effort over the past two decades is the development of language memory-models to deal with the conflict between language features and compiler/hardware optimizations, i.e., some language features are unsafe in the presence of aggressive sequential optimizations~\cite{Buhr95a,Boehm05}.
319The consequence is that a language must provide sufficient tools to program around safety issues, as inline and library code is all sequential to the compiler.
320One solution is low-level qualifiers and functions (e.g., @volatile@ and atomics) allowing \emph{programmers} to explicitly write safe (race-free~\cite{Boehm12}) programs.
321A safer solution is high-level language constructs so the \emph{compiler} knows the optimization boundaries, and hence, provides implicit safety.
322This problem is best know with respect to concurrency, but applies to other complex control-flow, like exceptions\footnote{
323\CFA exception handling will be presented in a separate paper.
324The key feature that dovetails with this paper is non-local exceptions allowing exceptions to be raised across stacks, with synchronous exceptions raised among coroutines and asynchronous exceptions raised among threads, similar to that in \uC~\cite[\S~5]{uC++}
325} and coroutines.
326Finally, solutions in the language allows matching constructs with language paradigm, i.e., imperative and functional languages have different presentations of the same concept.
327
328Finally, it is important for a language to provide safety over performance \emph{as the default}, allowing careful reduction of safety for performance when necessary.
329Two concurrency violations of this philosophy are \emph{spurious wakeup} and \emph{barging}, i.e., random wakeup~\cite[\S~8]{Buhr05a} and signals-as-hints~\cite[\S~8]{Buhr05a}, where one is a consequence of the other, i.e., once there is spurious wakeup, signals-as-hints follows.
330However, spurious wakeup is \emph{not} a foundational concurrency property~\cite[\S~8]{Buhr05a}, it is a performance design choice.
331Similarly, signals-as-hints is also a performance decision.
332We argue removing spurious wakeup and signals-as-hints makes concurrent programming significantly safer because it removes local non-determinism and matches with programmer expectation.
333(Experience by the authors teaching concurrency is that students are highly confused by these semantics.)
334Clawing back performance, when local non-determinism is unimportant, should be an option not the default.
335
336\begin{comment}
337For example, it is possible to provide exceptions, coroutines, monitors, and tasks as specialized types in an object-oriented language, integrating these constructs to allow leveraging the type-system (static type-checking) and all other object-oriented capabilities~\cite{uC++}.
338It is also possible to leverage call/return for blocking communication via new control structures, versus switching to alternative communication paradigms, like channels or message passing.
339As well, user threading is often a complementary feature, allowing light-weight threading to match with low-cost objects, while hiding the application/kernel boundary.
340User threading also allows layering of implicit concurrency models (no explicit thread creation), such executors, data-flow, actors, into a single language, so programmers can chose the model that best fits an algorithm.\footnote{
341All implicit concurrency models have explicit threading in their implementation, and hence, can be build from explicit threading;
342however, the reverse is seldom true, i.e., given implicit concurrency, e.g., actors, it is virtually impossible to create explicit concurrency, e.g., blocking thread objects.}
343Finally, with extended language features and user-level threading it is possible to discretely fold locking and non-blocking I/O multiplexing into the language's I/O libraries, so threading implicitly dovetails with the I/O subsystem.
344\CFA embraces language extensions and user-level threading to provide advanced control-flow (exception handling\footnote{
345\CFA exception handling will be presented in a separate paper.
346The key feature that dovetails with this paper is non-local 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++}
347} and coroutines) and concurrency.
348
349Most augmented traditional (Fortran 18~\cite{Fortran18}, Cobol 14~\cite{Cobol14}, Ada 12~\cite{Ada12}, Java 11~\cite{Java11}) and new languages (Go~\cite{Go}, Rust~\cite{Rust}, and D~\cite{D}), except \CC, diverge from C with different syntax and semantics, only interoperate indirectly with C, and are not systems languages, for those with managed memory.
350As a result, there is a significant learning curve to move to these languages, and C legacy-code must be rewritten.
351While \CC, like \CFA, takes an evolutionary approach to extend C, \CC's constantly growing complex and interdependent features-set (e.g., objects, inheritance, templates, etc.) mean idiomatic \CC code is difficult to use from C, and C programmers must expend significant effort learning \CC.
352Hence, rewriting and retraining costs for these languages, even \CC, are prohibitive for companies with a large C software-base.
353\CFA with its orthogonal feature-set, its high-performance runtime, and direct access to all existing C libraries circumvents these problems.
354\end{comment}
355
356\CFA embraces user-level threading, language extensions for advanced control-flow, and safety as the default.
357We present comparative examples so the reader can judge if the \CFA control-flow extensions are better and safer than those in or proposed for \Celeven, \CC, and other concurrent, imperative programming-languages, and perform experiments to show the \CFA runtime is competitive with other similar mechanisms.
358The main contributions of this work are:
359\begin{itemize}
360\item
361language-level generators, coroutines and user-level threading, which respect the expectations of C programmers.
362\item
363monitor synchronization without barging, and the ability to safely acquiring multiple monitors \emph{simultaneously} (deadlock free), while seamlessly integrating these capability with all monitor synchronization mechanisms.
364\item
365providing statically type-safe interfaces that integrate with the \CFA polymorphic type-system and other language features.
366% \item
367% library extensions for executors, futures, and actors built on the basic mechanisms.
368\item
369a runtime system with no spurious wakeup.
370\item
371a dynamic partitioning mechanism to segregate the execution environment for specialized requirements.
372% \item
373% a non-blocking I/O library
374\item
375experimental results showing comparable performance of the new features with similar mechanisms in other programming languages.
376\end{itemize}
377
378
379\section{Stateful Function}
380
381The generator/coroutine provides a stateful function, which is an old idea~\cite{Conway63,Marlin80} that is new again~\cite{C++20Coroutine19}.
382A stateful function allows execution to be temporarily suspended and later resumed, e.g., plugin, device driver, finite-state machine.
383Hence, 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.
384This capability is accomplished by retaining a data/execution \emph{closure} between invocations.
385If the closure is fixed size, we call it a \emph{generator} (or \emph{stackless}), and its control flow is restricted, e.g., suspending outside the generator is prohibited.
386If the closure is variable sized, we call it a \emph{coroutine} (or \emph{stackful}), and as the names implies, often implemented with a separate stack with no programming restrictions.
387Hence, refactoring a stackless coroutine may require changing it to stackful.
388A foundational property of all \emph{stateful functions} is that resume/suspend \emph{do not} cause incremental stack growth, i.e., resume/suspend operations are remembered through the closure not the stack.
389As well, activating a stateful function is \emph{asymmetric} or \emph{symmetric}, identified by resume/suspend (no cycles) and resume/resume (cycles).
390A fixed closure activated by modified call/return is faster than a variable closure activated by context switching.
391Additionally, any storage management for the closure (especially in unmanaged languages, i.e., no garbage collection) must also be factored into design and performance.
392Therefore, selecting between stackless and stackful semantics is a tradeoff between programming requirements and performance, where stackless is faster and stackful is more general.
393Note, creation cost is amortized across usage, so activation cost is usually the dominant factor.
394
395\begin{figure}
396\centering
397\begin{lrbox}{\myboxA}
398\begin{cfa}[aboveskip=0pt,belowskip=0pt]
399typedef struct {
400        int fn1, fn;
401} Fib;
402#define FibCtor { 1, 0 }
403int fib( Fib * f ) {
404
405
406
407        int fn = f->fn; f->fn = f->fn1;
408                f->fn1 = f->fn + fn;
409        return fn;
410
411}
412int main() {
413        Fib f1 = FibCtor, f2 = FibCtor;
414        for ( int i = 0; i < 10; i += 1 )
415                printf( "%d %d\n",
416                           fib( &f1 ), fib( &f2 ) );
417}
418\end{cfa}
419\end{lrbox}
420
421\begin{lrbox}{\myboxB}
422\begin{cfa}[aboveskip=0pt,belowskip=0pt]
423`generator` Fib {
424        int fn1, fn;
425};
426
427void `main(Fib & fib)` with(fib) {
428
429        [fn1, fn] = [1, 0];
430        for () {
431                `suspend;`
432                [fn1, fn] = [fn, fn + fn1];
433
434        }
435}
436int main() {
437        Fib f1, f2;
438        for ( 10 )
439                sout | `resume( f1 )`.fn
440                         | `resume( f2 )`.fn;
441}
442\end{cfa}
443\end{lrbox}
444
445\begin{lrbox}{\myboxC}
446\begin{cfa}[aboveskip=0pt,belowskip=0pt]
447typedef struct {
448        int fn1, fn;  void * `next`;
449} Fib;
450#define FibCtor { 1, 0, NULL }
451Fib * comain( Fib * f ) {
452        if ( f->next ) goto *f->next;
453        f->next = &&s1;
454        for ( ;; ) {
455                return f;
456          s1:; int fn = f->fn + f->fn1;
457                        f->fn1 = f->fn; f->fn = fn;
458        }
459}
460int main() {
461        Fib f1 = FibCtor, f2 = FibCtor;
462        for ( int i = 0; i < 10; i += 1 )
463                printf("%d %d\n",comain(&f1)->fn,
464                                 comain(&f2)->fn);
465}
466\end{cfa}
467\end{lrbox}
468
469\subfloat[C asymmetric generator]{\label{f:CFibonacci}\usebox\myboxA}
470\hspace{3pt}
471\vrule
472\hspace{3pt}
473\subfloat[\CFA asymmetric generator]{\label{f:CFAFibonacciGen}\usebox\myboxB}
474\hspace{3pt}
475\vrule
476\hspace{3pt}
477\subfloat[C generator implementation]{\label{f:CFibonacciSim}\usebox\myboxC}
478\caption{Fibonacci (output) asymmetric generator}
479\label{f:FibonacciAsymmetricGenerator}
480
481\bigskip
482
483\begin{lrbox}{\myboxA}
484\begin{cfa}[aboveskip=0pt,belowskip=0pt]
485`generator Fmt` {
486        char ch;
487        int g, b;
488};
489void ?{}( Fmt & fmt ) { `resume(fmt);` } // constructor
490void ^?{}( Fmt & f ) with(f) { $\C[1.75in]{// destructor}$
491        if ( g != 0 || b != 0 ) sout | nl; }
492void `main( Fmt & f )` with(f) {
493        for () { $\C{// until destructor call}$
494                for ( ; g < 5; g += 1 ) { $\C{// groups}$
495                        for ( ; b < 4; b += 1 ) { $\C{// blocks}$
496                                `suspend;` $\C{// wait for character}$
497                                while ( ch == '\n' ) `suspend;` // ignore
498                                sout | ch;                                              // newline
499                        } sout | " ";  // block spacer
500                } sout | nl; // group newline
501        }
502}
503int main() {
504        Fmt fmt; $\C{// fmt constructor called}$
505        for () {
506                sin | fmt.ch; $\C{// read into generator}$
507          if ( eof( sin ) ) break;
508                `resume( fmt );`
509        }
510
511} $\C{// fmt destructor called}\CRT$
512\end{cfa}
513\end{lrbox}
514
515\begin{lrbox}{\myboxB}
516\begin{cfa}[aboveskip=0pt,belowskip=0pt]
517typedef struct {
518        void * next;
519        char ch;
520        int g, b;
521} Fmt;
522void comain( Fmt * f ) {
523        if ( f->next ) goto *f->next;
524        f->next = &&s1;
525        for ( ;; ) {
526                for ( f->g = 0; f->g < 5; f->g += 1 ) {
527                        for ( f->b = 0; f->b < 4; f->b += 1 ) {
528                                return;
529                          s1:;  while ( f->ch == '\n' ) return;
530                                printf( "%c", f->ch );
531                        } printf( " " );
532                } printf( "\n" );
533        }
534}
535int main() {
536        Fmt fmt = { NULL };  comain( &fmt ); // prime
537        for ( ;; ) {
538                scanf( "%c", &fmt.ch );
539          if ( feof( stdin ) ) break;
540                comain( &fmt );
541        }
542        if ( fmt.g != 0 || fmt.b != 0 ) printf( "\n" );
543}
544\end{cfa}
545\end{lrbox}
546
547\subfloat[\CFA asymmetric generator]{\label{f:CFAFormatGen}\usebox\myboxA}
548\hspace{3pt}
549\vrule
550\hspace{3pt}
551\subfloat[C generator simulation]{\label{f:CFormatSim}\usebox\myboxB}
552\hspace{3pt}
553\caption{Formatter (input) asymmetric generator}
554\label{f:FormatterAsymmetricGenerator}
555\end{figure}
556
557For generators, coroutines, and threads, many designs are based on function objects or pointers~\cite{Butenhof97, C++14, MS:VisualC++, BoostCoroutines15}.
558For example, Python presents generators as a function object:
559\begin{python}
560def Gen():
561        ... `yield val` ...
562gen = Gen()
563for i in range( 10 ):
564        print( next( gen ) )
565\end{python}
566Boost presents coroutines in terms of four functor object-types:
567\begin{cfa}
568asymmetric_coroutine<>::pull_type
569asymmetric_coroutine<>::push_type
570symmetric_coroutine<>::call_type
571symmetric_coroutine<>::yield_type
572\end{cfa}
573and many languages present threading using function pointers, @pthreads@~\cite{Butenhof97}, \Csharp~\cite{Csharp}, Go~\cite{Go}, and Scala~\cite{Scala}, \eg pthreads:
574\begin{cfa}
575void * rtn( void * arg ) { ... }
576int i = 3, rc;
577pthread_t t; $\C{// thread id}$
578`rc = pthread_create( &t, rtn, (void *)i );` $\C{// create and initialized task, type-unsafe input parameter}$
579\end{cfa}
580% void mycor( pthread_t cid, void * arg ) {
581%       int * value = (int *)arg;                               $\C{// type unsafe, pointer-size only}$
582%       // thread body
583% }
584% int main() {
585%       int input = 0, output;
586%       coroutine_t cid = coroutine_create( &mycor, (void *)&input ); $\C{// type unsafe, pointer-size only}$
587%       coroutine_resume( cid, (void *)input, (void **)&output ); $\C{// type unsafe, pointer-size only}$
588% }
589\CFA's preferred presentation model for generators/coroutines/threads is a hybrid of objects and functions, with an object-oriented flavour.
590Essentially, the generator/coroutine/thread function is semantically coupled with a generator/coroutine/thread custom type.
591The custom type solves several issues, while accessing the underlying mechanisms used by the custom types is still allowed.
592
593
594\subsection{Generator}
595
596Stackless generators have the potential to be very small and fast, \ie as small and fast as function call/return for both creation and execution.
597The \CFA goal is to achieve this performance target, possibly at the cost of some semantic complexity.
598A series of different kinds of generators and their implementation demonstrate how this goal is accomplished.
599
600Figure~\ref{f:FibonacciAsymmetricGenerator} shows an unbounded asymmetric generator for an infinite sequence of Fibonacci numbers written in C and \CFA, with a simple C implementation for the \CFA version.
601This kind of generator is an \emph{output generator}, producing a new result on each resumption.
602To 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.
603An additional requirement is the ability to create an arbitrary number of generators (of any kind), \ie retaining state in global variables is insufficient;
604hence, state is retained in a closure between calls.
605Figure~\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.
606The C version only has the middle execution state because the top execution state becomes declaration initialization.
607Figure~\ref{f:CFAFibonacciGen} shows the \CFA approach, which also has a manual closure, but replaces the structure with a custom \CFA @generator@ type.
608This generator type is then connected to a function that \emph{must be named \lstinline|main|},\footnote{
609The name \lstinline|main| has special meaning in C, specifically the function where a program starts execution.
610Hence, overloading this name for other starting points (generator/coroutine/thread) is a logical extension.}
611which takes as its only parameter a reference to the generator type, called a \emph{generator main}.
612The generator main contains @suspend@ statements that suspend execution without ending the generator versus @return@.
613For the Fibonacci generator-main,\footnote{
614The \CFA \lstinline|with| opens an aggregate scope making its fields directly accessible, like Pascal \lstinline|with|, but using parallel semantics.
615Multiple aggregates may be opened.}
616the top initialization state appears at the start and the middle execution state is denoted by statement @suspend@.
617Any local variables in @main@ \emph{are not retained} between calls;
618hence local variable are only for temporary computations \emph{between} suspends.
619All retained state \emph{must} appear in the generator's type.
620As 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.
621The 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.
622Resuming an ended (returned) generator is undefined.
623Function @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.
624Figure~\ref{f:CFibonacciSim} shows the C implementation of the \CFA generator only needs one additional field, @next@, to handle retention of execution state.
625The computed @goto@ at the start of the generator main, which branches after the previous suspend, adds very little cost to the resume call.
626Finally, an explicit generator type provides both design and performance benefits, such as multiple type-safe interface functions taking and returning arbitrary types.
627\begin{cfa}
628int ?()( Fib & fib ) with( fib ) { return `resume( fib )`.fn; }   // function-call interface
629int ?()( Fib & fib, int N ) with( fib ) { for ( N - 1 ) `fib()`; return `fib()`; }   // use simple interface
630double ?()( Fib & fib ) with( fib ) { return (int)`fib()` / 3.14159; } // cast prevents recursive call
631sout | (int)f1() | (double)f1() | f2( 2 );   // simple interface, cast selects call based on return type, step 2 values
632\end{cfa}
633Now, the generator can be a separately-compiled opaque-type only accessed through its interface functions.
634For 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.
635
636Having to manually create the generator closure by moving local-state variables into the generator type is an additional programmer burden.
637(This restriction is removed by the coroutine, see Section~\ref{s:Coroutine}.)
638This requirement follows from the generality of variable-size local-state, \eg local state with a variable-length array requires dynamic allocation because the array size is unknown at compile time.
639However, dynamic allocation significantly increases the cost of generator creation/destruction and is a show-stopper for embedded real-time programming.
640But 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.
641With respect to safety, we believe static analysis can discriminate local state from temporary variables in a generator, \ie variable usage spanning @suspend@, and generate a compile-time error.
642Finally, our current experience is that most generator problems have simple data state, including local state, but complex execution state, so the burden of creating the generator type is small.
643As well, C programmers are not afraid with this kind of semantic programming requirement, if it results in very small, fast generators.
644
645Figure~\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.
646\begin{center}
647\tt
648\begin{tabular}{@{}l|l@{}}
649\multicolumn{1}{c|}{\textbf{\textrm{input}}} & \multicolumn{1}{c}{\textbf{\textrm{output}}} \\
650\begin{tabular}[t]{@{}ll@{}}
651abcdefghijklmnopqrstuvwxyz \\
652abcdefghijklmnopqrstuvwxyz
653\end{tabular}
654&
655\begin{tabular}[t]{@{}lllll@{}}
656abcd    & efgh  & ijkl  & mnop  & qrst  \\
657uvwx    & yzab  & cdef  & ghij  & klmn  \\
658opqr    & stuv  & wxyz  &               &
659\end{tabular}
660\end{tabular}
661\end{center}
662The 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.
663The destructor provides a newline, if formatted text ends with a full line.
664Figure~\ref{f:CFormatSim} shows the C implementation of the \CFA input generator with one additional field and the computed @goto@.
665For contrast, Figure~\ref{f:PythonFormatter} shows the equivalent Python format generator with the same properties as the Fibonacci generator.
666
667Figure~\ref{f:DeviceDriverGen} shows a \emph{killer} asymmetric generator, a device-driver, because device drivers caused 70\%-85\% of failures in Windows/Linux~\cite{Swift05}.
668Device drives follow the pattern of simple data state but complex execution state, \ie finite state-machine (FSM) parsing a protocol.
669For example, the following protocol:
670\begin{center}
671\ldots\, STX \ldots\, message \ldots\, ESC ETX \ldots\, message \ldots\, ETX 2-byte crc \ldots
672\end{center}
673is a network message beginning with the control character STX, ending with an ETX, and followed by a 2-byte cyclic-redundancy check.
674Control characters may appear in a message if preceded by an ESC.
675When 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.
676The device driver returns a status code of its current state, and when a complete message is obtained, the operating system knows the message is in the message buffer.
677Hence, the device driver is an input/output generator.
678
679Note, the cost of creating and resuming the device-driver generator, @Driver@, is virtually identical to call/return, so performance in an operating-system kernel is excellent.
680As well, the data state is small, where variables @byte@ and @msg@ are communication variables for passing in message bytes and returning the message, and variables @lnth@, @crc@, and @sum@ are local variable that must be retained between calls and are hoisted into the generator type.
681Manually, detecting and hoisting local-state variables is easy when the number is small.
682Finally, the execution state is large, with one @resume@ and seven @suspend@s.
683Hence, the key benefits of the generator are correctness, safety, and maintenance because the execution states of the FSM are transcribed directly into the programming language rather than using a table-driven approach.
684Because FSMs can be complex and occur frequently in important domains, direct support of the generator is crucial in a systems programming-language.
685
686\begin{figure}
687\centering
688\newbox\myboxA
689\begin{lrbox}{\myboxA}
690\begin{python}[aboveskip=0pt,belowskip=0pt]
691def Fib():
692        fn1, fn = 0, 1
693        while True:
694                `yield fn1`
695                fn1, fn = fn, fn1 + fn
696f1 = Fib()
697f2 = Fib()
698for i in range( 10 ):
699        print( next( f1 ), next( f2 ) )
700
701
702
703
704
705
706\end{python}
707\end{lrbox}
708
709\newbox\myboxB
710\begin{lrbox}{\myboxB}
711\begin{python}[aboveskip=0pt,belowskip=0pt]
712def Fmt():
713        try:
714                while True:
715                        for g in range( 5 ):
716                                for b in range( 4 ):
717                                        print( `(yield)`, end='' )
718                                print( '  ', end='' )
719                        print()
720        except GeneratorExit:
721                if g != 0 | b != 0:
722                        print()
723fmt = Fmt()
724`next( fmt )`                    # prime, next prewritten
725for i in range( 41 ):
726        `fmt.send( 'a' );`      # send to yield
727\end{python}
728\end{lrbox}
729\subfloat[Fibonacci]{\label{f:PythonFibonacci}\usebox\myboxA}
730\hspace{3pt}
731\vrule
732\hspace{3pt}
733\subfloat[Formatter]{\label{f:PythonFormatter}\usebox\myboxB}
734\caption{Python generator}
735\label{f:PythonGenerator}
736
737\bigskip
738
739\begin{tabular}{@{}l|l@{}}
740\begin{cfa}[aboveskip=0pt,belowskip=0pt]
741enum Status { CONT, MSG, ESTX,
742                                ELNTH, ECRC };
743`generator` Driver {
744        Status status;
745        unsigned char byte, * msg; // communication
746        unsigned int lnth, sum;      // local state
747        unsigned short int crc;
748};
749void ?{}( Driver & d, char * m ) { d.msg = m; }
750Status next( Driver & d, char b ) with( d ) {
751        byte = b; `resume( d );` return status;
752}
753void main( Driver & d ) with( d ) {
754        enum { STX = '\002', ESC = '\033',
755                        ETX = '\003', MaxMsg = 64 };
756  msg: for () { // parse message
757                status = CONT;
758                lnth = 0; sum = 0;
759                while ( byte != STX ) `suspend;`
760          emsg: for () {
761                        `suspend;` // process byte
762\end{cfa}
763&
764\begin{cfa}[aboveskip=0pt,belowskip=0pt]
765                        choose ( byte ) { // switch with implicit break
766                          case STX:
767                                status = ESTX; `suspend;` continue msg;
768                          case ETX:
769                                break emsg;
770                          case ESC:
771                                `suspend;`
772                        }
773                        if ( lnth >= MaxMsg ) { // buffer full ?
774                                status = ELNTH; `suspend;` continue msg; }
775                        msg[lnth++] = byte;
776                        sum += byte;
777                }
778                msg[lnth] = '\0'; // terminate string
779                `suspend;`
780                crc = byte << 8;
781                `suspend;`
782                status = (crc | byte) == sum ? MSG : ECRC;
783                `suspend;`
784        }
785}
786\end{cfa}
787\end{tabular}
788\caption{Device-driver generator for communication protocol}
789\label{f:DeviceDriverGen}
790\end{figure}
791
792Figure~\ref{f:CFAPingPongGen} shows a symmetric generator, where the generator resumes another generator, forming a resume/resume cycle.
793(The trivial cycle is a generator resuming itself.)
794This control flow is similar to recursion for functions, but without stack growth.
795The steps for symmetric control-flow are creating, executing, and terminating the cycle.
796Constructing 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.
797(This issues occurs for any cyclic data-structure.)
798% The example creates all the generators and then assigns the partners that form the cycle.
799% 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.
800Once the cycle is formed, the program main resumes one of the generators, and the generators can then traverse an arbitrary cycle using @resume@ to activate partner generator(s).
801Terminating 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).
802The starting stack-frame is below the last active generator because the resume/resume cycle does not grow the stack.
803Also, since local variables are not retained in the generator function, it does not contain an objects with destructors that must be called, so the  cost is the same as a function return.
804Destructor cost occurs when the generator instance is deallocated, which is easily controlled by the programmer.
805
806Figure~\ref{f:CPingPongSim} shows the implementation of the symmetric generator, where the complexity is the @resume@, which needs an extension to the calling convention to perform a forward rather than backward jump.
807This jump starts at the top of the next generator main to re-execute the normal calling convention to make space on the stack for its local variables.
808However, before the jump, the caller must reset its stack (and any registers) equivalent to a @return@, but subsequently jump forward not backwards.
809This semantics is basically a tail-call optimization, which compilers already perform.
810Hence, assembly code is manually insert in the example to undo the generator's entry code before the direct jump.
811This assembly code is fragile as it depends on what entry code is generated, specifically if there are local variables and the level of optimization.
812To provide this new calling convention requires a mechanism built into the compiler, which is beyond the scope of \CFA at this time.
813Nevertheless, it is possible to hand generate any symmetric generators for proof of concept and performance testing.
814A compiler could also eliminate other artifacts in the generator simulation to further increase performance.
815
816\begin{figure}
817\centering
818\begin{lrbox}{\myboxA}
819\begin{cfa}[aboveskip=0pt,belowskip=0pt]
820`generator PingPong` {
821        const char * name;
822        int N;
823        int i;                          // local state
824        PingPong & partner; // rebindable reference
825};
826
827void `main( PingPong & pp )` with(pp) {
828        for ( ; i < N; i += 1 ) {
829                sout | name | i;
830                `resume( partner );`
831        }
832}
833int main() {
834        enum { N = 5 };
835        PingPong ping = {"ping",N,0}, pong = {"pong",N,0};
836        &ping.partner = &pong;  &pong.partner = &ping;
837        `resume( ping );`
838}
839\end{cfa}
840\end{lrbox}
841
842\begin{lrbox}{\myboxB}
843\begin{cfa}[escapechar={},aboveskip=0pt,belowskip=0pt]
844typedef struct PingPong {
845        const char * name;
846        int N, i;
847        struct PingPong * partner;
848        void * next;
849} PingPong;
850#define PPCtor(name, N) {name,N,0,NULL,NULL}
851void comain( PingPong * pp ) {
852        if ( pp->next ) goto *pp->next;
853        pp->next = &&cycle;
854        for ( ; pp->i < pp->N; pp->i += 1 ) {
855                printf( "%s %d\n", pp->name, pp->i );
856                asm( "mov  %0,%%rdi" : "=m" (pp->partner) );
857                asm( "mov  %rdi,%rax" );
858                asm( "popq %rbx" );
859                asm( "jmp  comain" );
860          cycle: ;
861        }
862}
863\end{cfa}
864\end{lrbox}
865
866\subfloat[\CFA symmetric generator]{\label{f:CFAPingPongGen}\usebox\myboxA}
867\hspace{3pt}
868\vrule
869\hspace{3pt}
870\subfloat[C generator simulation]{\label{f:CPingPongSim}\usebox\myboxB}
871\hspace{3pt}
872\caption{Ping-Pong symmetric generator}
873\label{f:PingPongSymmetricGenerator}
874\end{figure}
875
876Finally, part of this generator work was inspired by the recent \CCtwenty generator proposal~\cite{C++20Coroutine19} (which they call coroutines).
877Our work provides the same high-performance asymmetric-generators as \CCtwenty, and extends their work with symmetric generators.
878An 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 the following \CFA syntax.
879\begin{cfa}
880... suspend`{ ... }`;
881... resume( C )`{ ... }` ...
882\end{cfa}
883Since the current generator's stack is released before calling the compound statement, the compound statement can only reference variables in the generator's type.
884This 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.
885Hence, this mechanism provides a general and safe handoff of the generator among competing threads.
886
887
888\subsection{Coroutine}
889\label{s:Coroutine}
890
891Stackful coroutines extend generator semantics, \ie there is an implicit closure and @suspend@ may appear in a helper function called from the coroutine main.
892A coroutine is specified by replacing @generator@ with @coroutine@ for the type.
893This generality results in higher cost for creation, due to dynamic stack allocation, execution, due to context switching among stacks, and terminating, due to possible stack unwinding and dynamic stack deallocation.
894A series of different kinds of coroutines and their implementation demonstrate how coroutines extend generators.
895
896First, 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.
897\begin{description}
898\item[Fibonacci]
899Move the declaration of @fn1@ to the start of coroutine main.
900\begin{cfa}[xleftmargin=0pt]
901void main( Fib & fib ) with(fib) {
902        `int fn1;`
903\end{cfa}
904\item[Formatter]
905Move the declaration of @g@ and @b@ to the for loops in the coroutine main.
906\begin{cfa}[xleftmargin=0pt]
907for ( `g`; 5 ) {
908        for ( `b`; 4 ) {
909\end{cfa}
910\item[Device Driver]
911Move the declaration of @lnth@ and @sum@ to their points of initialization.
912\begin{cfa}[xleftmargin=0pt]
913        status = CONT;
914        `unsigned int lnth = 0, sum = 0;`
915        ...
916        `unsigned short int crc = byte << 8;`
917\end{cfa}
918\item[PingPong]
919Move the declaration of @i@ to the for loop in the coroutine main.
920\begin{cfa}[xleftmargin=0pt]
921void main( PingPong & pp ) with(pp) {
922        for ( `i`; N ) {
923\end{cfa}
924\end{description}
925It 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.
926\begin{cfa}
927unsigned int Crc() {
928        `suspend;`
929        unsigned short int crc = byte << 8;
930        `suspend;`
931        status = (crc | byte) == sum ? MSG : ECRC;
932        return crc;
933}
934\end{cfa}
935A call to this function is placed at the end of the driver's coroutine-main.
936For complex finite-state machines, refactoring is part of normal program abstraction, especially when code is used in multiple places.
937Again, this complexity is usually associated with execution state rather than data state.
938
939\begin{comment}
940Figure~\ref{f:Coroutine3States} creates a @coroutine@ type, @`coroutine` Fib { int fn; }@, which provides communication, @fn@, for the \newterm{coroutine main}, @main@, which runs on the coroutine stack, and possibly multiple interface functions, \eg @next@.
941Like 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.
942The 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@.
943The interface function @next@, takes a Fibonacci instance and context switches to it using @resume@;
944on restart, the Fibonacci field, @fn@, contains the next value in the sequence, which is returned.
945The first @resume@ is special because it allocates the coroutine stack and cocalls its coroutine main on that stack;
946when the coroutine main returns, its stack is deallocated.
947Hence, @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.
948Figure~\ref{f:Coroutine1State} shows the coroutine version of the C version in Figure~\ref{f:ExternalState}.
949Coroutine generators are called \newterm{output coroutines} because values are only returned.
950
951\begin{figure}
952\centering
953\newbox\myboxA
954% \begin{lrbox}{\myboxA}
955% \begin{cfa}[aboveskip=0pt,belowskip=0pt]
956% `int fn1, fn2, state = 1;`   // single global variables
957% int fib() {
958%       int fn;
959%       `switch ( state )` {  // explicit execution state
960%         case 1: fn = 0;  fn1 = fn;  state = 2;  break;
961%         case 2: fn = 1;  fn2 = fn1;  fn1 = fn;  state = 3;  break;
962%         case 3: fn = fn1 + fn2;  fn2 = fn1;  fn1 = fn;  break;
963%       }
964%       return fn;
965% }
966% int main() {
967%
968%       for ( int i = 0; i < 10; i += 1 ) {
969%               printf( "%d\n", fib() );
970%       }
971% }
972% \end{cfa}
973% \end{lrbox}
974\begin{lrbox}{\myboxA}
975\begin{cfa}[aboveskip=0pt,belowskip=0pt]
976#define FibCtor { 0, 1 }
977typedef struct { int fn1, fn; } Fib;
978int fib( Fib * f ) {
979
980        int ret = f->fn1;
981        f->fn1 = f->fn;
982        f->fn = ret + f->fn;
983        return ret;
984}
985
986
987
988int main() {
989        Fib f1 = FibCtor, f2 = FibCtor;
990        for ( int i = 0; i < 10; i += 1 ) {
991                printf( "%d %d\n",
992                                fib( &f1 ), fib( &f2 ) );
993        }
994}
995\end{cfa}
996\end{lrbox}
997
998\newbox\myboxB
999\begin{lrbox}{\myboxB}
1000\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1001`coroutine` Fib { int fn1; };
1002void main( Fib & fib ) with( fib ) {
1003        int fn;
1004        [fn1, fn] = [0, 1];
1005        for () {
1006                `suspend;`
1007                [fn1, fn] = [fn, fn1 + fn];
1008        }
1009}
1010int ?()( Fib & fib ) with( fib ) {
1011        return `resume( fib )`.fn1;
1012}
1013int main() {
1014        Fib f1, f2;
1015        for ( 10 ) {
1016                sout | f1() | f2();
1017}
1018
1019
1020\end{cfa}
1021\end{lrbox}
1022
1023\newbox\myboxC
1024\begin{lrbox}{\myboxC}
1025\begin{python}[aboveskip=0pt,belowskip=0pt]
1026
1027def Fib():
1028
1029        fn1, fn = 0, 1
1030        while True:
1031                `yield fn1`
1032                fn1, fn = fn, fn1 + fn
1033
1034
1035// next prewritten
1036
1037
1038f1 = Fib()
1039f2 = Fib()
1040for i in range( 10 ):
1041        print( next( f1 ), next( f2 ) )
1042
1043
1044
1045\end{python}
1046\end{lrbox}
1047
1048\subfloat[C]{\label{f:GlobalVariables}\usebox\myboxA}
1049\hspace{3pt}
1050\vrule
1051\hspace{3pt}
1052\subfloat[\CFA]{\label{f:ExternalState}\usebox\myboxB}
1053\hspace{3pt}
1054\vrule
1055\hspace{3pt}
1056\subfloat[Python]{\label{f:ExternalState}\usebox\myboxC}
1057\caption{Fibonacci generator}
1058\label{f:C-fibonacci}
1059\end{figure}
1060
1061\bigskip
1062
1063\newbox\myboxA
1064\begin{lrbox}{\myboxA}
1065\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1066`coroutine` Fib { int fn; };
1067void main( Fib & fib ) with( fib ) {
1068        fn = 0;  int fn1 = fn; `suspend`;
1069        fn = 1;  int fn2 = fn1;  fn1 = fn; `suspend`;
1070        for () {
1071                fn = fn1 + fn2; fn2 = fn1; fn1 = fn; `suspend`; }
1072}
1073int next( Fib & fib ) with( fib ) { `resume( fib );` return fn; }
1074int main() {
1075        Fib f1, f2;
1076        for ( 10 )
1077                sout | next( f1 ) | next( f2 );
1078}
1079\end{cfa}
1080\end{lrbox}
1081\newbox\myboxB
1082\begin{lrbox}{\myboxB}
1083\begin{python}[aboveskip=0pt,belowskip=0pt]
1084
1085def Fibonacci():
1086        fn = 0; fn1 = fn; `yield fn`  # suspend
1087        fn = 1; fn2 = fn1; fn1 = fn; `yield fn`
1088        while True:
1089                fn = fn1 + fn2; fn2 = fn1; fn1 = fn; `yield fn`
1090
1091
1092f1 = Fibonacci()
1093f2 = Fibonacci()
1094for i in range( 10 ):
1095        print( `next( f1 )`, `next( f2 )` ) # resume
1096
1097\end{python}
1098\end{lrbox}
1099\subfloat[\CFA]{\label{f:Coroutine3States}\usebox\myboxA}
1100\qquad
1101\subfloat[Python]{\label{f:Coroutine1State}\usebox\myboxB}
1102\caption{Fibonacci input coroutine, 3 states, internal variables}
1103\label{f:cfa-fibonacci}
1104\end{figure}
1105\end{comment}
1106
1107\begin{figure}
1108\centering
1109\lstset{language=CFA,escapechar={},moredelim=**[is][\protect\color{red}]{`}{`}}% allow $
1110\begin{tabular}{@{}l@{\hspace{2\parindentlnth}}l@{}}
1111\begin{cfa}
1112`coroutine` Prod {
1113        Cons & c;                       // communication
1114        int N, money, receipt;
1115};
1116void main( Prod & prod ) with( prod ) {
1117        // 1st resume starts here
1118        for ( i; N ) {
1119                int p1 = random( 100 ), p2 = random( 100 );
1120                sout | p1 | " " | p2;
1121                int status = delivery( c, p1, p2 );
1122                sout | " $" | money | nl | status;
1123                receipt += 1;
1124        }
1125        stop( c );
1126        sout | "prod stops";
1127}
1128int payment( Prod & prod, int money ) {
1129        prod.money = money;
1130        `resume( prod );`
1131        return prod.receipt;
1132}
1133void start( Prod & prod, int N, Cons &c ) {
1134        &prod.c = &c;
1135        prod.[N, receipt] = [N, 0];
1136        `resume( prod );`
1137}
1138int main() {
1139        Prod prod;
1140        Cons cons = { prod };
1141        start( prod, 5, cons );
1142}
1143\end{cfa}
1144&
1145\begin{cfa}
1146`coroutine` Cons {
1147        Prod & p;                       // communication
1148        int p1, p2, status;
1149        bool done;
1150};
1151void ?{}( Cons & cons, Prod & p ) {
1152        &cons.p = &p; // reassignable reference
1153        cons.[status, done ] = [0, false];
1154}
1155void main( Cons & cons ) with( cons ) {
1156        // 1st resume starts here
1157        int money = 1, receipt;
1158        for ( ; ! done; ) {
1159                sout | p1 | " " | p2 | nl | " $" | money;
1160                status += 1;
1161                receipt = payment( p, money );
1162                sout | " #" | receipt;
1163                money += 1;
1164        }
1165        sout | "cons stops";
1166}
1167int delivery( Cons & cons, int p1, int p2 ) {
1168        cons.[p1, p2] = [p1, p2];
1169        `resume( cons );`
1170        return cons.status;
1171}
1172void stop( Cons & cons ) {
1173        cons.done = true;
1174        `resume( cons );`
1175}
1176
1177\end{cfa}
1178\end{tabular}
1179\caption{Producer / consumer: resume-resume cycle, bi-directional communication}
1180\label{f:ProdCons}
1181\end{figure}
1182
1183Figure~\ref{f:ProdCons} shows the ping-pong example in Figure~\ref{f:CFAPingPongGen} extended into a producer/consumer symmetric-coroutine performing bidirectional communication.
1184This example is illustrative because both producer/consumer have two interface functions with @resume@s that suspend execution in these interface (helper) functions.
1185The 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.
1186The first @resume@ of @prod@ creates @prod@'s stack with a frame for @prod@'s coroutine main at the top, and context switches to it.
1187@prod@'s coroutine main starts, creates local-state variables that are retained between coroutine activations, and executes $N$ iterations, each generating two random values, calling the consumer to deliver the values, and printing the status returned from the consumer.
1188
1189The producer call to @delivery@ transfers values into the consumer's communication variables, resumes the consumer, and returns the consumer status.
1190On the first resume, @cons@'s stack is created and initialized, holding local-state variables retained between subsequent activations of the coroutine.
1191The consumer iterates until the @done@ flag is set, prints the values delivered by the producer, increments status, and calls back to the producer via @payment@, and on return from @payment@, prints the receipt from the producer and increments @money@ (inflation).
1192The call from the consumer to @payment@ introduces the cycle between producer and consumer.
1193When @payment@ is called, the consumer copies values into the producer's communication variable and a resume is executed.
1194The context switch restarts the producer at the point where it last context switched, so it continues in @delivery@ after the resume.
1195@delivery@ returns the status value in @prod@'s coroutine main, where the status is printed.
1196The loop then repeats calling @delivery@, where each call resumes the consumer coroutine.
1197The context switch to the consumer continues in @payment@.
1198The consumer increments and returns the receipt to the call in @cons@'s coroutine main.
1199The loop then repeats calling @payment@, where each call resumes the producer coroutine.
1200Figure~\ref{f:ProdConsRuntimeStacks} shows the runtime stacks of the program main, and the coroutine mains for @prod@ and @cons@ during the cycling.
1201
1202\begin{figure}
1203\begin{center}
1204\input{FullProdConsStack.pstex_t}
1205\end{center}
1206\vspace*{-10pt}
1207\caption{Producer / consumer runtime stacks}
1208\label{f:ProdConsRuntimeStacks}
1209
1210\medskip
1211
1212\begin{center}
1213\input{FullCoroutinePhases.pstex_t}
1214\end{center}
1215\vspace*{-10pt}
1216\caption{Ping / Pong coroutine steps}
1217\label{f:PingPongFullCoroutineSteps}
1218\end{figure}
1219
1220Terminating 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.
1221Furthermore, each deallocated coroutine must guarantee all destructors are run for object allocated in the coroutine type \emph{and} allocated on the coroutine's stack at the point of suspension, which can be arbitrarily deep.
1222When a coroutine's main ends, its stack is already unwound so any stack allocated objects with destructors have been finalized.
1223The na\"{i}ve semantics for coroutine-cycle termination is context switch to the last resumer, like executing a @suspend@/@return@ in a generator.
1224However, for coroutines, the last resumer is \emph{not} implicitly below the current stack frame, as for generators, because each coroutine's stack is independent.
1225Unfortunately, it is impossible to determine statically if a coroutine is in a cycle and unrealistic to check dynamically (graph-cycle problem).
1226Hence, a compromise solution is necessary that works for asymmetric (acyclic) and symmetric (cyclic) coroutines.
1227
1228Our solution for coroutine termination works well for the most common asymmetric and symmetric coroutine usage-patterns.
1229For asymmetric coroutines, it is common for the first resumer (starter) coroutine to be the only resumer.
1230All previous generators converted to coroutines have this property.
1231For symmetric coroutines, it is common for the cycle creator to persist for the life-time of the cycle.
1232Hence, the starter coroutine is remembered on the first resume and ending the coroutine resumes the starter.
1233Figure~\ref{f:ProdConsRuntimeStacks} shows this semantic by the dashed lines from the end of the coroutine mains: @prod@ starts @cons@ so @cons@ resumes @prod@ at the end, and the program main starts @prod@ so @prod@ resumes the program main at the end.
1234For other scenarios, it is always possible to devise a solution with additional programming effort.
1235
1236The producer/consumer example does not illustrate the full power of the starter semantics because @cons@ always ends first.
1237Assume generator @PingPong@ is converted to a coroutine.
1238Figure~\ref{f:PingPongFullCoroutineSteps} shows the creation, starter, and cyclic execution steps of the coroutine version.
1239The program main creates (declares) coroutine instances @ping@ and @pong@.
1240Next, program main resumes @ping@, making it @ping@'s starter, and @ping@'s main resumes @pong@'s main, making it @pong@'s starter.
1241Execution forms a cycle when @pong@ resumes @ping@, and cycles $N$ times.
1242By adjusting $N$ for either @ping@/@pong@, it is possible to have either one finish first, instead of @pong@ always ending first.
1243If @pong@ ends first, it resumes its starter @ping@ in its coroutine main, then @ping@ ends and resumes its starter the program main in function @start@.
1244If @ping@ ends first, it resumes its starter the program main in function @start@.
1245Regardless of the cycle complexity, the starter stack always leads back to the program main, but the stack can be entered at an arbitrary point.
1246Once back at the program main, coroutines @ping@ and @pong@ are deallocated.
1247For generators, deallocation runs the destructors for all objects in the generator type.
1248For coroutines, deallocation deals with objects in the coroutine type and must also run the destructors for any objects pending on the coroutine's stack for any unterminated coroutine.
1249Hence, if a coroutine's destructor detects the coroutine is not ended, it implicitly raises a cancellation exception (uncatchable exception) at the coroutine and resumes it so the cancellation exception can propagate to the root of the coroutine's stack destroying all local variable on the stack.
1250So the \CFA semantics for the generator and coroutine, ensure both can be safely deallocated at any time, regardless of their current state, like any other aggregate object.
1251Explicitly raising normal exceptions at another coroutine can replace flag variables, like @stop@, \eg @prod@ raises a @stop@ exception at @cons@ after it finishes generating values and resumes @cons@, which catches the @stop@exception to terminate its loop.
1252
1253Finally, there is an interesting effect for @suspend@ with symmetric coroutines.
1254A coroutine must retain its last resumer to suspend back because the resumer is on a different stack.
1255These reverse pointers allow @suspend@ to cycle \emph{backwards}, which may be useful in certain cases.
1256However, there is an anomaly if a coroutine resumes itself, because it overwrites its last resumer with itself, losing the ability to resume the last external resumer.
1257To prevent losing this information, a self-resume does not overwrite the last resumer.
1258
1259
1260\subsection{(Generator) Coroutine Implementation}
1261
1262A significant implementation challenge for generators/coroutines (and threads, see Section~\ref{s:threads}) is adding extra fields to the custom types and related functions, \eg inserting code after/before the coroutine constructor/destructor and @main@ to create/initialize/de-initialize/destroy any extra fields, \eg stack.
1263There are several solutions to these problem, which follow from the object-oriented-flavoured choice to adopt custom types.
1264
1265For object-oriented languages, inheritance is used to provide extra fields and code via explicit inheritance:
1266\begin{cfa}[morekeywords={class,inherits}]
1267class myCoroutine inherits baseCoroutine { ... }
1268\end{cfa}
1269The 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.
1270As well, some special properties are not handled by existing language semantics, \eg the execution of constructors/destructors is in the wrong order to implicitly start threads because the thread must start \emph{after} all constructors as it relies on a completely initialized object, but the inherited constructor runs \emph{before} the derived.
1271Alternatives, such as explicitly starting threads as in Java, are repetitive and forgetting to call start is a common source of errors.
1272
1273An alternative is composition:
1274\begin{cfa}
1275struct myCoroutine {
1276        ... // declaration/communication variables
1277        baseCoroutine dummy; // composition, last declaration
1278}
1279\end{cfa}
1280which also requires an explicit declaration that must be last to ensure correct initialization order.
1281However, there is nothing preventing wrong placement or multiple declarations.
1282
1283\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 and when it is unsafe to perform certain optimizations, and IDEs using simple parsing can find and manipulate types with special properties.
1284The downside of this approach is that it makes custom types a special case in the language.
1285Users wanting to extend custom types or build their own can only do so in ways offered by the language.
1286Furthermore, implementing custom types without language supports may display the power of a programming language.
1287\CFA blends the two approaches, providing custom type for idiomatic \CFA code, extending and building new custom types is still possible, similar to Java approach with builtin and library concurrency.
1288
1289Part of the mechanism to generalize custom types is the \CFA trait, \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.
1290\begin{cfa}
1291trait is_coroutine( `dtype` T ) {
1292        void main( T & );
1293        coroutine_desc * get_coroutine( T & );
1294};
1295forall( `dtype` T | is_coroutine(T) ) void $suspend$( T & );
1296forall( `dtype` T | is_coroutine(T) ) void resume( T & );
1297\end{cfa}
1298Note, copying generators/coroutines/threads is not meaningful.
1299For example, a coroutine retains its last resumer and suspends back to it;
1300having a copy also suspend back to the same resumer is undefined semantics.
1301Furthermore, two coroutines cannot logically execute on the same stack.
1302A deep coroutine copy, which copies the stack, is also meaningless in an unmanaged language (no garbage collection), like C, because the stack may contain pointers to object within it that require updating for the copy.
1303Now, the @dtype@ property provides no \emph{implicit} copying operations and the @is_coroutine@ trait provides no \emph{explicit} copying operations, so all coroutines must be passed by reference (pointer).
1304The function definitions ensures there is a statically-typed @main@ function that is the starting point (first stack frame) of a coroutine, and a mechanism to get (read) the currently executing coroutine handle.
1305The @main@ function has no return value or additional parameters because the coroutine type allows an arbitrary number of interface functions with corresponding arbitrary typed input/output values versus fixed ones.
1306The 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 @suspend@ and @resume@.
1307
1308The \CFA custom-type @coroutine@ implicitly implements the getter and forward declarations required for implementing the coroutine main:
1309\begin{cquote}
1310\begin{tabular}{@{}ccc@{}}
1311\begin{cfa}
1312coroutine MyCor {
1313        int value;
1314
1315};
1316\end{cfa}
1317&
1318{\Large $\Rightarrow$}
1319&
1320\begin{tabular}{@{}ccc@{}}
1321\begin{cfa}
1322struct MyCor {
1323        int value;
1324        coroutine_desc cor;
1325};
1326\end{cfa}
1327&
1328\begin{cfa}
1329static inline coroutine_desc *
1330get_coroutine( MyCor & this ) {
1331        return &this.cor;
1332}
1333\end{cfa}
1334&
1335\begin{cfa}
1336void main( MyCor * this );
1337
1338
1339
1340\end{cfa}
1341\end{tabular}
1342\end{tabular}
1343\end{cquote}
1344The combination of custom types and fundamental @trait@ description of these types allows an concise specification for programmers and tools, while more advanced programmers can have tighter control over memory layout and initialization.
1345
1346Figure~\ref{f:CoroutineMemoryLayout} shows different memory-layout options for a coroutine (where a task is similar).
1347The coroutine handle is the @coroutine@ instance containing programmer specified global/communication variables across interface functions.
1348The coroutine descriptor contains all implicit declarations needed by the runtime, \eg @suspend@/@resume@, and can be part of the coroutine handle or separate.
1349The coroutine stack can appear in a number of locations and fixed or variable sized.
1350Hence, the coroutine's stack could be a VLS\footnote{
1351We are examining variable-sized structures (VLS), where fields can be variable-sized structures or arrays.
1352Once allocated, a VLS is fixed sized.}
1353on the allocating stack, provided the allocating stack is large enough.
1354For a VLS stack allocation, allocation/deallocation is an inexpensive adjustment of the stack point, modulo any stack constructor costs (\eg initial frame setup).
1355For heap stack allocation, allocation/deallocation is an expensive heap allocation (where the heap can be a shared resource), modulo any stack constructor costs.
1356With heap stack allocation, it is also possible to use a split (segmented) stack calling-convention, available with gcc and clang, so the stack is variable sized.
1357Currently, \CFA supports stack/heap allocated descriptors but only fixed-sized heap allocated stacks;
1358In \CFA debug-mode, the fixed-sized stack is terminated with a write-only page, which catches most stack overflows.
1359Experience teaching concurrency with \uC~\cite{CS343} shows fixed-sized stacks are rarely an issue for students.
1360Split-stack allocation is under development but requires recompilation of existing code, which may be impossible.
1361
1362\begin{figure}
1363\centering
1364\input{corlayout.pstex_t}
1365\caption{Coroutine memory layout}
1366\label{f:CoroutineMemoryLayout}
1367\end{figure}
1368
1369
1370\section{Concurrency}
1371\label{s:Concurrency}
1372
1373Concurrency is nondeterministic scheduling of independent sequential execution-paths (threads), where each thread has its own stack.
1374A single thread with multiple call stacks, \newterm{coroutining}~\cite{Conway63,Marlin80}, does \emph{not} imply concurrency~\cite[\S~2]{Buhr05a}.
1375In coroutining, coroutines self-schedule the thread across stacks so execution is deterministic.
1376(It is \emph{impossible} to generate a concurrency error when coroutining.)
1377However, coroutines are a stepping stone towards concurrency.
1378
1379The 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}.
1380Therefore, a minimal concurrency system requires coroutines \emph{in conjunction with a nondeterministic scheduler}.
1381The resulting execution system now follows a cooperative threading-model, called \newterm{non-preemptive scheduling}.
1382Adding \newterm{preemption} introduces non-cooperative scheduling, where context switching occurs randomly between any two instructions often based on a timer interrupt, called \newterm{preemptive scheduling}.
1383While a scheduler introduces uncertain execution among explicit context switches, preemption introduces uncertainty by introducing implicit context switches.
1384Uncertainty gives the illusion of parallelism on a single processor and provides a mechanism to access and increase performance on multiple processors.
1385The reason is that the scheduler/runtime have complete knowledge about resources and how to best utilized them.
1386However, the introduction of unrestricted nondeterminism results in the need for \newterm{mutual exclusion} and \newterm{synchronization}, which restrict nondeterminism for correctness;
1387otherwise, it is impossible to write meaningful concurrent programs.
1388Optimal concurrent performance is often obtained by having as much nondeterminism as mutual exclusion and synchronization correctness allow.
1389
1390A scheduler can either be a stackless or stackful.
1391For 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.
1392For 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.
1393A stackful scheduler is often used for simplicity and security.
1394
1395
1396\subsection{Thread}
1397\label{s:threads}
1398
1399Threading needs the ability to start a thread and wait for its completion.
1400A common API for this ability is @fork@ and @join@.
1401\begin{cquote}
1402\begin{tabular}{@{}lll@{}}
1403\multicolumn{1}{c}{\textbf{Java}} & \multicolumn{1}{c}{\textbf{\Celeven}} & \multicolumn{1}{c}{\textbf{pthreads}} \\
1404\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1405class MyTask extends Thread {...}
1406mytask t = new MyTask(...);
1407`t.start();` // start
1408// concurrency
1409`t.join();` // wait
1410\end{cfa}
1411&
1412\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1413class MyTask { ... } // functor
1414MyTask mytask;
1415`thread t( mytask, ... );` // start
1416// concurrency
1417`t.join();` // wait
1418\end{cfa}
1419&
1420\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1421void * rtn( void * arg ) {...}
1422pthread_t t;  int i = 3;
1423`pthread_create( &t, rtn, (void *)i );` // start
1424// concurrency
1425`pthread_join( t, NULL );` // wait
1426\end{cfa}
1427\end{tabular}
1428\end{cquote}
1429\CFA has a simpler approach using a custom @thread@ type and leveraging declaration semantics (allocation/deallocation), where threads implicitly @fork@ after construction and @join@ before destruction.
1430\begin{cfa}
1431thread MyTask {};
1432void main( MyTask & this ) { ... }
1433int main() {
1434        MyTask team`[10]`; $\C[2.5in]{// allocate stack-based threads, implicit start after construction}$
1435        // concurrency
1436} $\C{// deallocate stack-based threads, implicit joins before destruction}$
1437\end{cfa}
1438This semantic ensures a thread is started and stopped exactly once, eliminating some programming error, and scales to multiple threads for basic (termination) synchronization.
1439For block allocation to arbitrary depth, including recursion, threads are created/destroyed in a lattice structure (tree with top and bottom).
1440Arbitrary topologies are possible using dynamic allocation, allowing threads to outlive their declaration scope, identical to normal dynamically allocating.
1441\begin{cfa}
1442MyTask * factory( int N ) { ... return `anew( N )`; } $\C{// allocate heap-based threads, implicit start after construction}$
1443int main() {
1444        MyTask * team = factory( 10 );
1445        // concurrency
1446        `delete( team );` $\C{// deallocate heap-based threads, implicit joins before destruction}\CRT$
1447}
1448\end{cfa}
1449
1450Figure~\ref{s:ConcurrentMatrixSummation} shows concurrently adding the rows of a matrix and then totalling the subtotals sequentially, after all the row threads have terminated.
1451The program uses heap-based threads because each thread needs different constructor values.
1452(Python provides a simple iteration mechanism to initialize array elements to different values allowing stack allocation.)
1453The allocation/deallocation pattern appears unusual because allocated objects are immediately deallocated without any intervening code.
1454However, for threads, the deletion provides implicit synchronization, which is the intervening code.
1455% 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.
1456
1457\begin{figure}
1458\begin{cfa}
1459`thread` Adder { int * row, cols, & subtotal; } $\C{// communication variables}$
1460void ?{}( Adder & adder, int row[], int cols, int & subtotal ) {
1461        adder.[ row, cols, &subtotal ] = [ row, cols, &subtotal ];
1462}
1463void main( Adder & adder ) with( adder ) {
1464        subtotal = 0;
1465        for ( c; cols ) { subtotal += row[c]; }
1466}
1467int main() {
1468        const int rows = 10, cols = 1000;
1469        int matrix[rows][cols], subtotals[rows], total = 0;
1470        // read matrix
1471        Adder * adders[rows];
1472        for ( r; rows; ) { $\C{// start threads to sum rows}$
1473                adders[r] = `new( matrix[r], cols, &subtotals[r] );`
1474        }
1475        for ( r; rows ) { $\C{// wait for threads to finish}$
1476                `delete( adders[r] );` $\C{// termination join}$
1477                total += subtotals[r]; $\C{// total subtotal}$
1478        }
1479        sout | total;
1480}
1481\end{cfa}
1482\caption{Concurrent matrix summation}
1483\label{s:ConcurrentMatrixSummation}
1484\end{figure}
1485
1486
1487\subsection{Thread Implementation}
1488
1489Threads in \CFA are user level run by runtime kernel threads (see Section~\ref{s:Parallelism}), where user threads provide concurrency and kernel threads provide parallelism.
1490Like coroutines, and for the same design reasons, \CFA provides a custom @thread@ type and a @trait@ to enforce and restrict the task-interface functions.
1491\begin{cquote}
1492\begin{tabular}{@{}c@{\hspace{3\parindentlnth}}c@{}}
1493\begin{cfa}
1494thread myThread {
1495        ... // declaration/communication variables
1496};
1497
1498
1499\end{cfa}
1500&
1501\begin{cfa}
1502trait is_thread( `dtype` T ) {
1503        void main( T & );
1504        thread_desc * get_thread( T & );
1505        void ^?{}( T & `mutex` );
1506};
1507\end{cfa}
1508\end{tabular}
1509\end{cquote}
1510Like coroutines, the @dtype@ property prevents \emph{implicit} copy operations and the @is_coroutine@ trait provides no \emph{explicit} copy operations, so threads must be passed by reference (pointer).
1511Similarly, the function definitions ensures there is a statically-typed @main@ function that is the thread starting point (first stack frame), a mechanism to get (read) the currently executing thread handle, and a special destructor to prevent deallocation while the thread is executing.
1512(The qualifier @mutex@ for the destructor parameter is discussed in Section~\ref{s:Monitor}.)
1513The 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 starts running the coroutine main on the stack;
1514whereas, a thread is scheduling for execution in @main@ immediately after its constructor is run.
1515No return value or additional parameters are necessary for this function because the @thread@ type allows an arbitrary number of interface functions with corresponding arbitrary typed input/output values.
1516
1517\begin{comment} % put in appendix with coroutine version ???
1518As such the @main@ function of a thread can be defined as
1519\begin{cfa}
1520thread foo {};
1521
1522void main(foo & this) {
1523        sout | "Hello World!";
1524}
1525\end{cfa}
1526
1527In this example, threads of type @foo@ start execution in the @void main(foo &)@ function, which prints @"Hello World!".@ While this paper encourages this approach to enforce strongly typed programming, users may prefer to use the function-based thread semantics for the sake of simplicity.
1528With the static semantics it is trivial to write a thread type that takes a function pointer as a parameter and executes it on its stack asynchronously.
1529\begin{cfa}
1530typedef void (*voidRtn)(int);
1531
1532thread RtnRunner {
1533        voidRtn func;
1534        int arg;
1535};
1536
1537void ?{}(RtnRunner & this, voidRtn inRtn, int arg) {
1538        this.func = inRtn;
1539        this.arg  = arg;
1540}
1541
1542void main(RtnRunner & this) {
1543        // thread starts here and runs the function
1544        this.func( this.arg );
1545}
1546
1547void hello(/*unused*/ int) {
1548        sout | "Hello World!";
1549}
1550
1551int main() {
1552        RtnRunner f = {hello, 42};
1553        return 0?
1554}
1555\end{cfa}
1556A consequence of the strongly typed approach to main is that memory layout of parameters and return values to/from a thread are now explicitly specified in the \textbf{API}.
1557\end{comment}
1558
1559
1560\section{Mutual Exclusion / Synchronization}
1561
1562Uncontrolled nondeterministic execution produces meaningless results.
1563To 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}.
1564Some concurrent systems eliminate mutable shared-state by switching to stateless communication like message passing~\cite{Thoth,Harmony,V-Kernel,MPI} (Erlang, MPI), channels~\cite{CSP} (CSP,Go), actors~\cite{Akka} (Akka, Scala), or functional techniques (Haskell).
1565However, in call/return-based languages, these approaches force a clear distinction, \ie introduce a new programming paradigm between regular and concurrent computation, \eg function call versus message passing.
1566Hence, a programmer must learn and manipulate two sets of design/programming patterns.
1567While this distinction can be hidden away in library code, effective use of the library still has to take both paradigms into account.
1568In contrast, approaches based on stateful models more closely resemble the standard call/return programming-model, resulting in a single programming paradigm.
1569
1570At the lowest level, concurrent control is implemented by atomic operations, upon which different kinds of locking mechanisms are constructed, \eg semaphores~\cite{Dijkstra68b}, barriers, and path expressions~\cite{Campbell74}.
1571However, for productivity it is always desirable to use the highest-level construct that provides the necessary efficiency~\cite{Hochstein05}.
1572A newer approach for restricting non-determinism is transactional memory~\cite{Herlihy93}.
1573While this approach is pursued in hardware~\cite{Nakaike15} and system languages, like \CC~\cite{Cpp-Transactions}, the performance and feature set is still too restrictive to be the main concurrency paradigm for system languages, which is why it is rejected as the core paradigm for concurrency in \CFA.
1574
1575One of the most natural, elegant, and efficient mechanisms for mutual exclusion and synchronization for shared-memory systems is the \emph{monitor}.
1576First 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}.
1577In addition, operating-system kernels and device drivers have a monitor-like structure, although they often use lower-level primitives such as mutex locks or semaphores to simulate monitors.
1578For these reasons, \CFA selected monitors as the core high-level concurrency-construct, upon which higher-level approaches can be easily constructed.
1579
1580
1581\subsection{Mutual Exclusion}
1582
1583A group of instructions manipulating a specific instance of shared data that must be performed atomically is called an (individual) \newterm{critical-section}~\cite{Dijkstra65}.
1584The generalization is called a \newterm{group critical-section}~\cite{Joung00}, where multiple tasks with the same session may use the resource simultaneously, but different sessions may not use the resource simultaneously.
1585The readers/writer problem~\cite{Courtois71} is an instance of a group critical-section, where readers share a session but writers have a unique session.
1586\newterm{Mutual exclusion} enforces the correct kind and number of threads use a critical section.
1587
1588However, many solutions exist for mutual exclusion, which vary in terms of performance, flexibility and ease of use.
1589Methods range from low-level locks, which are fast and flexible but require significant attention for correctness, to higher-level concurrency techniques, which sacrifice some performance to improve ease of use.
1590Ease of use comes by either guaranteeing some problems cannot occur, \eg deadlock free, or by offering a more explicit coupling between shared data and critical section.
1591For example, the \CC @std::atomic<T>@ offers an easy way to express mutual-exclusion on a restricted set of operations, \eg reading/writing, for numerical types.
1592However, a significant challenge with locks is composability because it takes careful organization for multiple locks to be used while preventing deadlock.
1593Easing composability is another feature higher-level mutual-exclusion mechanisms can offer.
1594
1595
1596\subsection{Synchronization}
1597
1598Synchronization enforces relative ordering of execution, and synchronization tools provide numerous mechanisms to establish these timing relationships.
1599Low-level synchronization primitives offer good performance and flexibility at the cost of ease of use;
1600higher-level mechanisms often simplify usage by adding better coupling between synchronization and data, \eg receive-specific versus receive-any thread in message passing or offering specialized solutions, \eg barrier lock.
1601Often synchronization is used to order access to a critical section, \eg ensuring a waiting writer thread enters the critical section before a calling reader thread.
1602If the calling reader is scheduled before the waiting writer, the reader has \newterm{barged}.
1603Barging can result in staleness/freshness problems, where a reader barges ahead of a writer and reads temporally stale data, or a writer barges ahead of another writer overwriting data with a fresh value preventing the previous value from ever being read (lost computation).
1604Preventing or detecting barging is an involved challenge with low-level locks, which is made easier through higher-level constructs.
1605This challenge is often split into two different approaches: barging avoidance and prevention.
1606Algorithms that unconditionally releasing a lock for competing threads to acquire use barging avoidance during synchronization to force a barging thread to wait.
1607algorithms that conditionally hold locks during synchronization, \eg baton-passing~\cite{Andrews89}, prevent barging completely.
1608
1609
1610\section{Monitor}
1611\label{s:Monitor}
1612
1613A \textbf{monitor} is a set of functions that ensure mutual exclusion when accessing shared state.
1614More precisely, a monitor is a programming technique that binds mutual exclusion to function scope, as opposed to locks, where mutual-exclusion is defined by acquire/release calls, independent of lexical context (analogous to block and heap storage allocation).
1615The strong association with the call/return paradigm eases programming, comprehension, and maintenance, at a slight cost in flexibility and efficiency.
1616\CFA uses a custom @monitor@ type and leverages declaration semantics (deallocation) to protect active or waiting threads in a monitor.
1617
1618The following is a \CFA monitor implementation of an atomic counter.
1619\newpage
1620\begin{cfa}[morekeywords=nomutex]
1621`monitor` Aint { int cnt; }; $\C[4.25in]{// atomic integer counter}$
1622int ++?( Aint & `mutex`$\(_{opt}\)$ this ) with( this ) { return ++cnt; } $\C{// increment}$
1623int ?=?( Aint & `mutex`$\(_{opt}\)$ lhs, int rhs ) with( lhs ) { cnt = rhs; } $\C{// conversions}\CRT$
1624int ?=?( int & lhs, Aint & `mutex`$\(_{opt}\)$ rhs ) with( rhs ) { lhs = cnt; }
1625\end{cfa}
1626% The @Aint@ constructor, @?{}@, uses the \lstinline[morekeywords=nomutex]@nomutex@ qualifier indicating mutual exclusion is unnecessary during construction because an object is inaccessible (private) until after it is initialized.
1627% (While a constructor may publish its address into a global variable, doing so generates a race-condition.)
1628The prefix increment operation, @++?@, is normally @mutex@, indicating mutual exclusion is necessary during function execution, to protect the incrementing from race conditions, unless there is an atomic increment instruction for the implementation type.
1629The assignment operators provide bi-directional conversion between an atomic and normal integer without accessing field @cnt@;
1630these operations only need @mutex@, if reading/writing the implementation type is not atomic.
1631The atomic counter is used without any explicit mutual-exclusion and provides thread-safe semantics, which is similar to the \CC template @std::atomic@.
1632\begin{cfa}
1633int i = 0, j = 0, k = 5;
1634Aint x = { 0 }, y = { 0 }, z = { 5 }; $\C{// no mutex required}$
1635++x; ++y; ++z; $\C{// safe increment by multiple threads}$
1636x = 2; y = i; z = k; $\C{// conversions}$
1637i = x; j = y; k = z;
1638\end{cfa}
1639
1640\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.
1641\begin{cfa}
1642monitor M { ... } m;
1643void foo( M & mutex m ) { ... } $\C{// acquire mutual exclusion}$
1644void bar( M & mutex m ) { $\C{// acquire mutual exclusion}$
1645        ... `bar( m );` ... `foo( m );` ... $\C{// reacquire mutual exclusion}$
1646}
1647\end{cfa}
1648\CFA monitors also ensure the monitor lock is always 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.
1649Similar safety is offered by explicit mechanisms like \CC RAII;
1650monitor implicit safety ensures no programmer error.
1651Furthermore, RAII mechansims cannot handle complex synchronization within a monitor, where the monitor lock may not be released on function exit but passed to an unblocking thread.
1652RAII is purely a mutual-exclusion mechanism (see Section~\ref{s:Scheduling}).
1653
1654
1655\subsection{Monitor Implementation}
1656
1657For the same design reasons, \CFA provides a custom @monitor@ type and a @trait@ to enforce and restrict the monitor-interface functions.
1658\begin{cquote}
1659\begin{tabular}{@{}c@{\hspace{3\parindentlnth}}c@{}}
1660\begin{cfa}
1661monitor M {
1662        ... // shared data
1663};
1664
1665\end{cfa}
1666&
1667\begin{cfa}
1668trait is_monitor( `dtype` T ) {
1669        monitor_desc * get_monitor( T & );
1670        void ^?{}( T & mutex );
1671};
1672\end{cfa}
1673\end{tabular}
1674\end{cquote}
1675Like before, the @dtype@ property prevents \emph{implicit} copy operations and the @is_monitor@ trait provides no \emph{explicit} copy operations, so monitors must be passed by reference (pointer).
1676% Copying a lock is insecure because it is possible to copy an open lock and then use the open copy when the original lock is closed to simultaneously access the shared data.
1677% Copying a monitor is secure because both the lock and shared data are copies, but copying the shared data is meaningless because it no longer represents a unique entity.
1678Similarly, the function definitions ensures there is a mechanism to get (read) the currently executing monitor handle, and a special destructor to prevent deallocation if a thread using the shared data.
1679The custom monitor type also inserts any locking vehicles needed to implement the monitor locking semnatics.
1680
1681
1682\subsection{Mutex Acquisition}
1683\label{s:MutexAcquisition}
1684
1685While the monitor lock provides mutual exclusion for shared data, there are implementation options for when and where the locking/unlocking occurs.
1686(Much of this discussion also applies to basic locks.)
1687For example, a monitor may be passed through multiple helper functions before it is necessary to acquire the monitor's mutual exclusion.
1688
1689The benefit of mandatory monitor qualifiers is self-documentation, but requiring both @mutex@ and \lstinline[morekeywords=nomutex]@nomutex@ for all monitor parameters is redundant.
1690Instead, the semantics have one qualifier as the default, and the other required.
1691For example, make the safe @mutex@ qualifier the default because assuming \lstinline[morekeywords=nomutex]@nomutex@ may cause subtle errors.
1692Alternatively, make the unsafe \lstinline[morekeywords=nomutex]@nomutex@ qualifier the default because it is the \emph{normal} parameter semantics while @mutex@ parameters are rare.
1693Providing a default qualifier implies knowing whether a parameter is a monitor.
1694Since \CFA relies heavily on traits as an abstraction mechanism, the distinction between a type that is a monitor and a type that looks like a monitor can become blurred.
1695For this reason, \CFA requires programmers to identify the kind of parameter with the @mutex@ keyword and uses no keyword to mean \lstinline[morekeywords=nomutex]@nomutex@.
1696
1697The next semantic decision is establishing which parameter \emph{types} may be qualified with @mutex@.
1698The following has monitor parameter types that are composed of multiple objects.
1699\begin{cfa}
1700monitor M { ... }
1701int f1( M & mutex m ); $\C{// single parameter object}$
1702int f2( M * mutex m ); $\C{// single or multiple parameter object}$
1703int f3( M * mutex m[$\,$] ); $\C{// multiple parameter object}$
1704int f4( stack( M * ) & mutex m ); $\C{// multiple parameters object}$
1705\end{cfa}
1706Function @f1@ has a single parameter object, while @f2@'s indirection could be a single or multi-element array, where static array size is often unknown in C.
1707Function @f3@ has a multiple object matrix, and @f4@ a multiple object data structure.
1708While shown shortly, multiple object acquisition is possible, but the number of objects must be statically known.
1709Therefore, \CFA only acquires one monitor per parameter with at most one level of indirection, excluding pointers as it is impossible to statically determine the size.
1710
1711For object-oriented monitors, \eg Java, calling a mutex member \emph{implicitly} acquires mutual exclusion of the receiver object, @`rec`.foo(...)@.
1712\CFA has no receiver, and hence, the explicit @mutex@ qualifier is used to specify which objects acquire mutual exclusion.
1713A positive consequence of this design decision is the ability to support multi-monitor functions,\footnote{
1714While object-oriented monitors can be extended with a mutex qualifier for multiple-monitor members, no prior example of this feature could be found.}
1715called \newterm{bulk acquire} (see Section~\ref{s:Implementation} for implementation details).
1716\CFA guarantees acquisition order is consistent across calls to @mutex@ functions using the same monitors as arguments, so acquiring multiple monitors is safe from deadlock.
1717Figure~\ref{f:BankTransfer} shows a trivial solution to the bank-account transfer problem using \CFA monitors with implicit locking and \CC with explicit locking.
1718A \CFA programmer only has to manage when to acquire mutual exclusion;
1719a \CC programmer must select the correct lock and acquisition mechanism from a panoply of locking options.
1720Making good choices for common cases in \CFA simplifies the programming experience and enhances safety.
1721
1722\begin{figure}
1723\centering
1724\begin{lrbox}{\myboxA}
1725\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1726monitor BankAccount {
1727
1728        int balance;
1729} b1 = { 0 }, b2 = { 0 };
1730void deposit( BankAccount & `mutex` b,
1731                        int deposit ) with(b) {
1732        balance += deposit;
1733}
1734void transfer( BankAccount & `mutex` my,
1735        BankAccount & `mutex` your, int me2you ) {
1736
1737        deposit( my, -me2you ); // debit
1738        deposit( your, me2you ); // credit
1739}
1740`thread` Person { BankAccount & b1, & b2; };
1741void main( Person & person ) with(person) {
1742        for ( 10_000_000 ) {
1743                if ( random() % 3 ) deposit( b1, 3 );
1744                if ( random() % 3 ) transfer( b1, b2, 7 );
1745        }
1746}   
1747int main() {
1748        `Person p1 = { b1, b2 }, p2 = { b2, b1 };`
1749
1750} // wait for threads to complete
1751\end{cfa}
1752\end{lrbox}
1753
1754\begin{lrbox}{\myboxB}
1755\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1756struct BankAccount {
1757        `recursive_mutex m;`
1758        int balance = 0;
1759} b1, b2;
1760void deposit( BankAccount & b, int deposit ) {
1761        `scoped_lock lock( b.m );`
1762        b.balance += deposit;
1763}
1764void transfer( BankAccount & my,
1765                        BankAccount & your, int me2you ) {
1766        `scoped_lock lock( my.m, your.m );`
1767        deposit( my, -me2you ); // debit
1768        deposit( your, me2you ); // credit
1769}
1770
1771void person( BankAccount & b1, BankAccount & b2 ) {
1772        for ( int i = 0; i < 10$'$000$'$000; i += 1 ) {
1773                if ( random() % 3 ) deposit( b1, 3 );
1774                if ( random() % 3 ) transfer( b1, b2, 7 );
1775        }
1776}   
1777int main() {
1778        `thread p1(person, ref(b1), ref(b2)), p2(person, ref(b2), ref(b1));`
1779        `p1.join(); p2.join();`
1780}
1781\end{cfa}
1782\end{lrbox}
1783
1784\subfloat[\CFA]{\label{f:CFABank}\usebox\myboxA}
1785\hspace{3pt}
1786\vrule
1787\hspace{3pt}
1788\subfloat[\CC]{\label{f:C++Bank}\usebox\myboxB}
1789\hspace{3pt}
1790\caption{Bank transfer problem}
1791\label{f:BankTransfer}
1792\end{figure}
1793
1794Users can force the acquiring order, by using @mutex@/\lstinline[morekeywords=nomutex]@nomutex@.
1795\begin{cfa}
1796void foo( M & mutex m1, M & mutex m2 ); $\C{// acquire m1 and m2}$
1797void bar( M & mutex m1, M & /* nomutex */ m2 ) { $\C{// acquire m1}$
1798        ... foo( m1, m2 ); ... $\C{// acquire m2}$
1799}
1800void baz( M & /* nomutex */ m1, M & mutex m2 ) { $\C{// acquire m2}$
1801        ... foo( m1, m2 ); ... $\C{// acquire m1}$
1802}
1803\end{cfa}
1804The bulk-acquire semantics allow @bar@ or @baz@ to acquire a monitor lock and reacquire it in @foo@.
1805In the calls to @bar@ and @baz@, the monitors are acquired in opposite order, resulting in deadlock.
1806However, this case is the simplest instance of the \emph{nested-monitor problem}~\cite{Lister77}, where monitors are acquired in sequence versus bulk.
1807Detecting the nested-monitor problem requires dynamic tracking of monitor calls, and dealing with it requires rollback semantics~\cite{Dice10}.
1808\CFA does not deal with this fundamental problem.
1809
1810
1811\subsection{\protect\lstinline@mutex@ statement}
1812\label{mutex-stmt}
1813
1814The monitor call-semantics associate all locking semantics with functions.
1815Like Java, \CFA offers an alternative @mutex@ statement to reduce refactoring and naming.
1816\begin{cquote}
1817\begin{tabular}{@{}l@{\hspace{3\parindentlnth}}l@{}}
1818\multicolumn{1}{c}{\textbf{\lstinline@mutex@ call}} & \multicolumn{1}{c}{\lstinline@mutex@ \textbf{statement}} \\
1819\begin{cfa}
1820monitor M { ... };
1821void foo( M & mutex m1, M & mutex m2 ) {
1822        // critical section
1823}
1824void bar( M & m1, M & m2 ) {
1825        foo( m1, m2 );
1826}
1827\end{cfa}
1828&
1829\begin{cfa}
1830
1831void bar( M & m1, M & m2 ) {
1832        mutex( m1, m2 ) {       // remove refactoring and naming
1833                // critical section
1834        }
1835}
1836
1837\end{cfa}
1838\end{tabular}
1839\end{cquote}
1840
1841
1842\subsection{Scheduling}
1843\label{s:Scheduling}
1844
1845While monitor mutual-exclusion provides safe access to shared data, the monitor data may indicate that a thread accessing it cannot proceed, \eg a bounded buffer may be full/empty so produce/consumer threads must block.
1846Leaving the monitor and trying again (busy waiting) is impractical for high-level programming.
1847Monitors eliminate busy waiting by providing synchronization to schedule threads needing access to the shared data, where threads block versus spinning.
1848Synchronization is generally achieved with internal~\cite{Hoare74} or external~\cite[\S~2.9.2]{uC++} scheduling, where \newterm{scheduling} defines which thread acquires the critical section next.
1849\newterm{Internal scheduling} is characterized by each thread entering the monitor and making an individual decision about proceeding or blocking, while \newterm{external scheduling} is characterized by an entering thread making a decision about proceeding for itself and on behalf of other threads attempting entry.
1850Finally, \CFA monitors do not allow calling threads to barge ahead of signalled threads, which simplifies synchronization among threads in the monitor and increases correctness.
1851If barging is allowed, synchronization between a signaller and signallee is difficult, often requiring additional flags and multiple unblock/block cycles.
1852In fact, signals-as-hints is completely opposite from that proposed by Hoare in the seminal paper on monitors:
1853\begin{cquote}
1854However, 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.
1855It is only in this way that a waiting program has an absolute guarantee that it can acquire the resource just released by the signalling program without any danger that a third program will interpose a monitor entry and seize the resource instead.~\cite[p.~550]{Hoare74}
1856\end{cquote}
1857Furthermore, \CFA concurrency has no spurious wakeup~\cite[\S~9]{Buhr05a}, which eliminates an implicit form of barging.
1858Hence, a \CFA @wait@ statement is not enclosed in a @while@ loop that can cause thread starvation.
1859
1860Figure~\ref{f:BBInt} shows a \CFA generic bounded-buffer with internal scheduling, where producers/consumers enter the monitor, see the buffer is full/empty, and block on an appropriate condition lock, @full@/@empty@.
1861The @wait@ function atomically blocks the calling thread and implicitly releases the monitor lock(s) for all monitors in the function's parameter list.
1862The appropriate condition lock is signalled to unblock an opposite kind of thread after an element is inserted/removed from the buffer.
1863Signalling is unconditional, because signalling an empty condition lock does nothing.
1864
1865Signalling semantics cannot have the signaller and signalled thread in the monitor simultaneously, which means:
1866\begin{enumerate}
1867\item
1868The signalling thread returns immediately, and the signalled thread continues.
1869\item
1870The signalling thread continues and the signalled thread is marked for urgent unblocking at the next scheduling point (exit/wait).
1871\item
1872The signalling thread blocks but is marked for urgrent unblocking at the next scheduling point and the signalled thread continues.
1873\end{enumerate}
1874The first approach is too restrictive, as it precludes solving a reasonable class of problems, \eg dating service (see Figure~\ref{f:DatingService}).
1875\CFA supports the next two semantics as both are useful.
1876Finally, while it is common to store a @condition@ as a field of the monitor, in \CFA, a @condition@ variable can be created/stored independently.
1877Furthermore, a condition variable is tied to a \emph{group} of monitors on first use, called \newterm{branding}, which means that using internal scheduling with distinct sets of monitors requires one condition variable per set of monitors.
1878
1879\begin{figure}
1880\centering
1881\newbox\myboxA
1882\begin{lrbox}{\myboxA}
1883\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1884forall( otype T ) { // distribute forall
1885        monitor Buffer {
1886                `condition` full, empty;
1887                int front, back, count;
1888                T elements[10];
1889        };
1890        void ?{}( Buffer(T) & buffer ) with(buffer) {
1891                [front, back, count] = 0;
1892        }
1893
1894        void insert( Buffer(T) & mutex buffer, T elem )
1895                                with(buffer) {
1896                if ( count == 10 ) `wait( empty )`;
1897                // insert elem into buffer
1898                `signal( full )`;
1899        }
1900        T remove( Buffer(T) & mutex buffer ) with(buffer) {
1901                if ( count == 0 ) `wait( full )`;
1902                // remove elem from buffer
1903                `signal( empty )`;
1904                return elem;
1905        }
1906}
1907\end{cfa}
1908\end{lrbox}
1909
1910\newbox\myboxB
1911\begin{lrbox}{\myboxB}
1912\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1913forall( otype T ) { // distribute forall
1914        monitor Buffer {
1915
1916                int front, back, count;
1917                T elements[10];
1918        };
1919        void ?{}( Buffer(T) & buffer ) with(buffer) {
1920                [front, back, count] = 0;
1921        }
1922        T remove( Buffer(T) & mutex buffer ); // forward
1923        void insert( Buffer(T) & mutex buffer, T elem )
1924                                with(buffer) {
1925                if ( count == 10 ) `waitfor( remove, buffer )`;
1926                // insert elem into buffer
1927
1928        }
1929        T remove( Buffer(T) & mutex buffer ) with(buffer) {
1930                if ( count == 0 ) `waitfor( insert, buffer )`;
1931                // remove elem from buffer
1932
1933                return elem;
1934        }
1935}
1936\end{cfa}
1937\end{lrbox}
1938
1939\subfloat[Internal scheduling]{\label{f:BBInt}\usebox\myboxA}
1940%\qquad
1941\subfloat[External scheduling]{\label{f:BBExt}\usebox\myboxB}
1942\caption{Generic bounded-buffer}
1943\label{f:GenericBoundedBuffer}
1944\end{figure}
1945
1946Figure~\ref{f:BBExt} shows a \CFA generic bounded-buffer with external scheduling, where producers/consumers detecting a full/\-empty buffer block and prevent more producers/consumers from entering the monitor until there is a free/empty slot in the buffer.
1947External scheduling is controlled by the @waitfor@ statement, which atomically blocks the calling thread, releases the monitor lock, and restricts the function calls that can next acquire mutual exclusion.
1948If the buffer is full, only calls to @remove@ can acquire the buffer, and if the buffer is empty, only calls to @insert@ can acquire the buffer.
1949Threads making calls to functions that are currently excluded, block outside of (external to) the monitor on a calling queue, versus blocking on condition queues inside of (internal to) the monitor.
1950External scheduling allows users to wait for events from other threads without concern of unrelated events occurring.
1951The mechnaism can be done in terms of control flow, \eg Ada @accept@ or \uC @_Accept@, or in terms of data, \eg Go @select@ on channels.
1952While both mechanisms have strengths and weaknesses, this project uses a control-flow mechanism to stay consistent with other language semantics.
1953Two challenges specific to \CFA for external scheduling are loose object-definitions (see Section~\ref{s:LooseObjectDefinitions}) and multiple-monitor functions (see Section~\ref{s:Multi-MonitorScheduling}).
1954
1955For internal scheduling, non-blocking signalling (as in the producer/consumer example) is used when the signaller is providing the cooperation for a waiting thread;
1956the signaller enters the monitor and changes state, detects a waiting threads that can use the state, performs a non-blocking signal on the condition queue for the waiting thread, and exits the monitor to run concurrently.
1957The waiter unblocks next from the urgent queue, uses/takes the state, and exits the monitor.
1958Blocking signalling is the reverse, where the waiter is providing the cooperation for the signalling thread;
1959the signaller enters the monitor, detects a waiting thread providing the necessary state, performs a blocking signal to place it on the urgent queue and unblock the waiter.
1960The waiter changes state and exits the monitor, and the signaller unblocks next from the urgent queue to use/take the state.
1961
1962Figure~\ref{f:DatingService} shows a dating service demonstrating non-blocking and blocking signalling.
1963The dating service matches girl and boy threads with matching compatibility codes so they can exchange phone numbers.
1964A thread blocks until an appropriate partner arrives.
1965The complexity is exchanging phone numbers in the monitor because of the mutual-exclusion property.
1966For 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.
1967For signal-block scheduling, the implicit urgent-queue replaces the explict @exchange@-condition and @signal_block@ puts the finding thread on the urgent condition and unblocks the matcher.
1968The dating service is an example of a monitor that cannot be written using external scheduling because it requires knowledge of calling parameters to make scheduling decisions, and parameters of waiting threads are unavailable;
1969as well, an arriving thread may not find a partner and must wait, which requires a condition variable, and condition variables imply internal scheduling.
1970Furthermore, 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.
1971Putting loops around the @wait@s does not correct the problem;
1972the solution must be restructured to account for barging.
1973
1974\begin{figure}
1975\centering
1976\newbox\myboxA
1977\begin{lrbox}{\myboxA}
1978\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1979enum { CCodes = 20 };
1980monitor DS {
1981        int GirlPhNo, BoyPhNo;
1982        condition Girls[CCodes], Boys[CCodes];
1983        condition exchange;
1984};
1985int girl( DS & mutex ds, int phNo, int ccode ) {
1986        if ( is_empty( Boys[ccode] ) ) {
1987                wait( Girls[ccode] );
1988                GirlPhNo = phNo;
1989                `signal( exchange );`
1990        } else {
1991                GirlPhNo = phNo;
1992                `signal( Boys[ccode] );`
1993                `wait( exchange );`
1994        } // if
1995        return BoyPhNo;
1996}
1997int boy( DS & mutex ds, int phNo, int ccode ) {
1998        // as above with boy/girl interchanged
1999}
2000\end{cfa}
2001\end{lrbox}
2002
2003\newbox\myboxB
2004\begin{lrbox}{\myboxB}
2005\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2006
2007monitor DS {
2008        int GirlPhNo, BoyPhNo;
2009        condition Girls[CCodes], Boys[CCodes];
2010
2011};
2012int girl( DS & mutex ds, int phNo, int ccode ) {
2013        if ( is_empty( Boys[ccode] ) ) { // no compatible
2014                wait( Girls[ccode] ); // wait for boy
2015                GirlPhNo = phNo; // make phone number available
2016
2017        } else {
2018                GirlPhNo = phNo; // make phone number available
2019                `signal_block( Boys[ccode] );` // restart boy
2020
2021        } // if
2022        return BoyPhNo;
2023}
2024int boy( DS & mutex ds, int phNo, int ccode ) {
2025        // as above with boy/girl interchanged
2026}
2027\end{cfa}
2028\end{lrbox}
2029
2030\subfloat[\lstinline@signal@]{\label{f:DatingSignal}\usebox\myboxA}
2031\qquad
2032\subfloat[\lstinline@signal_block@]{\label{f:DatingSignalBlock}\usebox\myboxB}
2033\caption{Dating service}
2034\label{f:DatingService}
2035\end{figure}
2036
2037Both internal and external scheduling extend to multiple monitors in a natural way.
2038\begin{cquote}
2039\begin{tabular}{@{}l@{\hspace{3\parindentlnth}}l@{}}
2040\begin{cfa}
2041monitor M { `condition e`; ... };
2042void foo( M & mutex m1, M & mutex m2 ) {
2043        ... wait( `e` ); ...   // wait( e, m1, m2 )
2044        ... wait( `e, m1` ); ...
2045        ... wait( `e, m2` ); ...
2046}
2047\end{cfa}
2048&
2049\begin{cfa}
2050void rtn$\(_1\)$( M & mutex m1, M & mutex m2 );
2051void rtn$\(_2\)$( M & mutex m1 );
2052void bar( M & mutex m1, M & mutex m2 ) {
2053        ... waitfor( `rtn` ); ...       // $\LstCommentStyle{waitfor( rtn\(_1\), m1, m2 )}$
2054        ... waitfor( `rtn, m1` ); ... // $\LstCommentStyle{waitfor( rtn\(_2\), m1 )}$
2055}
2056\end{cfa}
2057\end{tabular}
2058\end{cquote}
2059For @wait( e )@, the default semantics is to atomically block the signaller and release all acquired mutex types in the parameter list, \ie @wait( e, m1, m2 )@.
2060To override the implicit multi-monitor wait, specific mutex parameter(s) can be specified, \eg @wait( e, m1 )@.
2061Wait statically verifies the released monitors are the acquired mutex-parameters so unconditional release is safe.
2062While \CC supports bulk locking, @wait@ only accepts a single lock for a condition variable, so bulk locking with condition variables is asymmetric.
2063Finally, a signaller,
2064\begin{cfa}
2065void baz( M & mutex m1, M & mutex m2 ) {
2066        ... signal( e ); ...
2067}
2068\end{cfa}
2069must have acquired at least the same locks as the waiting thread signalled from the condition queue.
2070
2071Similarly, for @waitfor( rtn )@, the default semantics is to atomically block the acceptor and release all acquired mutex types in the parameter list, \ie @waitfor( rtn, m1, m2 )@.
2072To override the implicit multi-monitor wait, specific mutex parameter(s) can be specified, \eg @waitfor( rtn, m1 )@.
2073@waitfor@ statically verifies the released monitors are the same as the acquired mutex-parameters of the given function or function pointer.
2074To statically verify the released monitors match with the accepted function's mutex parameters, the function (pointer) prototype must be accessible.
2075% When an overloaded function appears in an @waitfor@ statement, calls to any function with that name are accepted.
2076% The rationale is that members with the same name should perform a similar function, and therefore, all should be eligible to accept a call.
2077Overloaded functions can be disambiguated using a cast
2078\begin{cfa}
2079void rtn( M & mutex m );
2080`int` rtn( M & mutex m );
2081waitfor( (`int` (*)( M & mutex ))rtn, m );
2082\end{cfa}
2083
2084The ability to release a subset of acquired monitors can result in a \newterm{nested monitor}~\cite{Lister77} deadlock.
2085\begin{cfa}
2086void foo( M & mutex m1, M & mutex m2 ) {
2087        ... wait( `e, m1` ); ...                                $\C{// release m1, keeping m2 acquired )}$
2088void bar( M & mutex m1, M & mutex m2 ) {        $\C{// must acquire m1 and m2 )}$
2089        ... signal( `e` ); ...
2090\end{cfa}
2091The @wait@ only releases @m1@ so the signalling thread cannot acquire @m1@ and @m2@ to enter @bar@ and @signal@ the condition.
2092While deadlock can occur with multiple/nesting acquisition, this is a consequence of locks, and by extension monitors, not being perfectly composable.
2093
2094% Supporting barging prevention as well as extending internal scheduling to multiple monitors is the main source of complexity in the design and implementation of \CFA concurrency.
2095
2096
2097\subsection{Barging Prevention}
2098
2099Figure~\ref{f:BargingPrevention} shows \CFA code where bulk acquire adds complexity to the internal-signalling semantics.
2100The complexity begins at the end of the inner @mutex@ statement, where the semantics of internal scheduling need to be extended for multiple monitors.
2101The problem is that bulk acquire is used in the inner @mutex@ statement where one of the monitors is already acquired.
2102When the signalling thread reaches the end of the inner @mutex@ statement, it should transfer ownership of @m1@ and @m2@ to the waiting threads to prevent barging into the outer @mutex@ statement by another thread.
2103However, both the signalling and waiting thread W1 still need monitor @m1@.
2104
2105\begin{figure}
2106\newbox\myboxA
2107\begin{lrbox}{\myboxA}
2108\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2109monitor M m1, m2;
2110condition c;
2111mutex( m1 ) { // $\LstCommentStyle{\color{red}outer}$
2112        ...
2113        mutex( m1, m2 ) { // $\LstCommentStyle{\color{red}inner}$
2114                ... `signal( c )`; ...
2115                // m1, m2 acquired
2116        } // $\LstCommentStyle{\color{red}release m2}$
2117        // m1 acquired
2118} // release m1
2119\end{cfa}
2120\end{lrbox}
2121
2122\newbox\myboxB
2123\begin{lrbox}{\myboxB}
2124\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2125
2126
2127mutex( m1 ) {
2128        ...
2129        mutex( m1, m2 ) {
2130                ... `wait( c )`; // block and release m1, m2
2131                // m1, m2 acquired
2132        } // $\LstCommentStyle{\color{red}release m2}$
2133        // m1 acquired
2134} // release m1
2135\end{cfa}
2136\end{lrbox}
2137
2138\newbox\myboxC
2139\begin{lrbox}{\myboxC}
2140\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2141
2142
2143mutex( m2 ) {
2144        ... `wait( c )`; ...
2145        // m2 acquired
2146} // $\LstCommentStyle{\color{red}release m2}$
2147
2148
2149
2150
2151\end{cfa}
2152\end{lrbox}
2153
2154\begin{cquote}
2155\subfloat[Signalling Thread]{\label{f:SignallingThread}\usebox\myboxA}
2156\hspace{3\parindentlnth}
2157\subfloat[Waiting Thread (W1)]{\label{f:WaitingThread}\usebox\myboxB}
2158\hspace{2\parindentlnth}
2159\subfloat[Waiting Thread (W2)]{\label{f:OtherWaitingThread}\usebox\myboxC}
2160\end{cquote}
2161\caption{Barging Prevention}
2162\label{f:BargingPrevention}
2163\end{figure}
2164
2165One scheduling solution is for the signaller to keep ownership of all locks until the last lock is ready to be transferred, because this semantics fits most closely to the behaviour of single-monitor scheduling.
2166However, Figure~\ref{f:OtherWaitingThread} shows this solution is complex depending on other waiters, resulting in options when the signaller finishes the inner mutex-statement.
2167The signaller can retain @m2@ until completion of the outer mutex statement and pass the locks to waiter W1, or it can pass @m2@ to waiter W2 after completing the inner mutex-statement, while continuing to hold @m1@.
2168In the latter case, waiter W2 must eventually pass @m2@ to waiter W1, which is complex because W1 may have waited before W2, so W2 is unaware of it.
2169Furthermore, there is an execution sequence where the signaller always finds waiter W2, and hence, waiter W1 starves.
2170
2171While a number of approaches were examined~\cite[\S~4.3]{Delisle18}, the solution chosen for \CFA is a novel techique called \newterm{partial signalling}.
2172Signalled threads are moved to the urgent queue and the waiter at the front defines the set of monitors necessary for it to unblock.
2173Partial signalling transfers ownership of monitors to the front waiter.
2174When the signaller thread exits or waits in the monitor, the front waiter is unblocked if all its monitors are released.
2175The 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.
2176
2177
2178\subsection{Loose Object Definitions}
2179\label{s:LooseObjectDefinitions}
2180
2181In an object-oriented programming-language, a class includes an exhaustive list of operations.
2182However, new members can be added via static inheritance or dynamic members, \eg JavaScript~\cite{JavaScript}.
2183Similarly, monitor functions can be added at any time in \CFA, making it less clear for programmers and more difficult to implement.
2184\begin{cfa}
2185monitor M { ... };
2186void `f`( M & mutex m );
2187void g( M & mutex m ) { waitfor( `f` ); } $\C{// clear which f}$
2188void `f`( M & mutex m, int ); $\C{// different f}$
2189void h( M & mutex m ) { waitfor( `f` ); } $\C{// unclear which f}$
2190\end{cfa}
2191Hence, the \CFA code for entering a monitor looks like:
2192\begin{cfa}
2193if ( $\textrm{\textit{monitor is free}}$ ) $\C{// enter}$
2194else if ( $\textrm{\textit{already own monitor}}$ ) $\C{// continue}$
2195else if ( $\textrm{\textit{monitor accepts me}}$ ) $\C{// enter}$
2196else $\C{// block}$
2197\end{cfa}
2198For the first two conditions, it is easy to implement a check that can evaluate the condition in a few instructions.
2199However, a fast check for \emph{monitor accepts me} is much harder to implement depending on the constraints put on the monitors.
2200Figure~\ref{fig:ClassicalMonitor} shows monitors are often expressed as an entry (calling) queue, some acceptor queues, and an urgent stack/queue.
2201
2202\begin{figure}
2203\centering
2204\subfloat[Classical monitor] {
2205\label{fig:ClassicalMonitor}
2206{\resizebox{0.45\textwidth}{!}{\input{monitor.pstex_t}}}
2207}% subfloat
2208\quad
2209\subfloat[Bulk acquire monitor] {
2210\label{fig:BulkMonitor}
2211{\resizebox{0.45\textwidth}{!}{\input{ext_monitor.pstex_t}}}
2212}% subfloat
2213\caption{Monitor implementation}
2214\label{f:MonitorImplementation}
2215\end{figure}
2216
2217For a fixed (small) number of mutex functions (\eg 128), the accept check reduces to a bitmask of allowed callers, which can be checked with a single instruction.
2218This approach requires a unique dense ordering of functions with a small upper-bound and the ordering must be consistent across translation units.
2219For object-oriented languages these constraints are common, but \CFA mutex functions can be added in any scope and are only visible in certain translation unit, precluding program-wide dense-ordering among mutex functions.
2220
2221Figure~\ref{fig:BulkMonitor} shows the \CFA monitor implementation.
2222The mutex function called is associated with each thread on the entry queue, while a list of acceptable functions is kept separately.
2223The accepted list is a variable-sized array of accepted function pointers, so the single instruction bitmask comparison is replaced by dereferencing a pointer followed by a (usually short) linear search.
2224
2225
2226\subsection{Multi-Monitor Scheduling}
2227\label{s:Multi-MonitorScheduling}
2228
2229External scheduling, like internal scheduling, becomes significantly more complex for multi-monitor semantics.
2230Even in the simplest case, new semantics needs to be established.
2231\begin{cfa}
2232monitor M { ... };
2233void f( M & mutex m1 );
2234void g( M & mutex m1, M & mutex m2 ) { `waitfor( f );` } $\C{// pass m1 or m2 to f?}$
2235\end{cfa}
2236The solution is for the programmer to disambiguate:
2237\begin{cfa}
2238waitfor( f, `m2` ); $\C{// wait for call to f with argument m2}$
2239\end{cfa}
2240Both 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@.
2241This behaviour can be extended to the multi-monitor @waitfor@ statement.
2242\begin{cfa}
2243monitor M { ... };
2244void f( M & mutex m1, M & mutex m2 );
2245void g( M & mutex m1, M & mutex m2 ) { waitfor( f, `m1, m2` ); $\C{// wait for call to f with arguments m1 and m2}$
2246\end{cfa}
2247Again, the set of monitors passed to the @waitfor@ statement must be entirely contained in the set of monitors already acquired by the accepting function.
2248Also, the order of the monitors in a @waitfor@ statement is unimportant.
2249
2250Figure~\ref{f:UnmatchedMutexSets} shows an example where, for internal and external scheduling with multiple monitors, a signalling or accepting thread must match exactly, \ie partial matching results in waiting.
2251For both examples, the set of monitors is disjoint so unblocking is impossible.
2252
2253\begin{figure}
2254\centering
2255\begin{lrbox}{\myboxA}
2256\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2257monitor M1 {} m11, m12;
2258monitor M2 {} m2;
2259condition c;
2260void f( M1 & mutex m1, M2 & mutex m2 ) {
2261        signal( c );
2262}
2263void g( M1 & mutex m1, M2 & mutex m2 ) {
2264        wait( c );
2265}
2266g( `m11`, m2 ); // block on wait
2267f( `m12`, m2 ); // cannot fulfil
2268\end{cfa}
2269\end{lrbox}
2270
2271\begin{lrbox}{\myboxB}
2272\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2273monitor M1 {} m11, m12;
2274monitor M2 {} m2;
2275
2276void f( M1 & mutex m1, M2 & mutex m2 ) {
2277
2278}
2279void g( M1 & mutex m1, M2 & mutex m2 ) {
2280        waitfor( f, m1, m2 );
2281}
2282g( `m11`, m2 ); // block on accept
2283f( `m12`, m2 ); // cannot fulfil
2284\end{cfa}
2285\end{lrbox}
2286\subfloat[Internal scheduling]{\label{f:InternalScheduling}\usebox\myboxA}
2287\hspace{3pt}
2288\vrule
2289\hspace{3pt}
2290\subfloat[External scheduling]{\label{f:ExternalScheduling}\usebox\myboxB}
2291\caption{Unmatched \protect\lstinline@mutex@ sets}
2292\label{f:UnmatchedMutexSets}
2293\end{figure}
2294
2295
2296\subsection{Extended \protect\lstinline@waitfor@}
2297
2298Figure~\ref{f:ExtendedWaitfor} show the extended form of the @waitfor@ statement to conditionally accept one of a group of mutex functions, with a specific action to be performed \emph{after} the mutex function finishes.
2299For a @waitfor@ clause to be executed, its @when@ must be true and an outstanding call to its corresponding member(s) must exist.
2300The \emph{conditional-expression} of a @when@ may call a function, but the function must not block or context switch.
2301If there are multiple acceptable mutex calls, selection occurs top-to-bottom (prioritized) in the @waitfor@ clauses, whereas some programming languages with similar mechanisms accept nondeterministically for this case, \eg Go \lstinline[morekeywords=select]@select@.
2302If some accept guards are true and there are no outstanding calls to these members, the acceptor is accept-blocked until a call to one of these members is made.
2303If 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.
2304Hence, the terminating @else@ clause allows a conditional attempt to accept a call without blocking.
2305If there is a @timeout@ clause, it provides an upper bound on waiting.
2306If both a @timeout@ clause and an @else@ clause are present, the @else@ must be conditional, or the @timeout@ is never triggered.
2307In all cases, the statement following is executed \emph{after} a clause is executed to know which of the clauses executed.
2308
2309\begin{figure}
2310\centering
2311\begin{cfa}
2312`when` ( $\emph{conditional-expression}$ )      $\C{// optional guard}$
2313        waitfor( $\emph{mutex-member-name}$ )
2314                $\emph{statement}$                                      $\C{// action after call}$
2315`or` `when` ( $\emph{conditional-expression}$ ) $\C{// optional guard}$
2316        waitfor( $\emph{mutex-member-name}$ )
2317                $\emph{statement}$                                      $\C{// action after call}$
2318`or`    ...                                                                     $\C{// list of waitfor clauses}$
2319`when` ( $\emph{conditional-expression}$ )      $\C{// optional guard}$
2320        `timeout`                                                               $\C{// optional terminating timeout clause}$
2321                $\emph{statement}$                                      $\C{// action after timeout}$
2322`when` ( $\emph{conditional-expression}$ )      $\C{// optional guard}$
2323        `else`                                                                  $\C{// optional terminating clause}$
2324                $\emph{statement}$                                      $\C{// action when no immediate calls}$
2325\end{cfa}
2326\caption{Extended \protect\lstinline@waitfor@}
2327\label{f:ExtendedWaitfor}
2328\end{figure}
2329
2330Note, a group of conditional @waitfor@ clauses is \emph{not} the same as a group of @if@ statements, e.g.:
2331\begin{cfa}
2332if ( C1 ) waitfor( mem1 );                       when ( C1 ) waitfor( mem1 );
2333else if ( C2 ) waitfor( mem2 );         or when ( C2 ) waitfor( mem2 );
2334\end{cfa}
2335The left example only accepts @mem1@ if @C1@ is true or only @mem2@ if @C2@ is true.
2336The right example accepts either @mem1@ or @mem2@ if @C1@ and @C2@ are true.
2337
2338An interesting use of @waitfor@ is accepting the @mutex@ destructor to know when an object is deallocated.
2339\begin{cfa}
2340void insert( Buffer(T) & mutex buffer, T elem ) with( buffer ) {
2341        if ( count == 10 )
2342                waitfor( remove, buffer ) {
2343                        // insert elem into buffer
2344                } or `waitfor( ^?{}, buffer )` throw insertFail;
2345}
2346\end{cfa}
2347When the buffer is deallocated, the current waiter is unblocked and informed, so it can perform an appropriate action.
2348However, the basic @waitfor@ semantics do not support this functionality, since using an object after its destructor is called is undefined.
2349Therefore, to make this useful capability work, the semantics for accepting the destructor is the same as @signal@, \ie the call to the destructor is placed on the urgent queue and the acceptor continues execution, which throws an exception to the acceptor and then the caller is unblocked from the urgent queue to deallocate the object.
2350Accepting the destructor is an idiomatic way to terminate a thread in \CFA.
2351
2352
2353\subsection{\protect\lstinline@mutex@ Threads}
2354
2355Threads in \CFA can also be monitors to allow \emph{direct communication} among threads, \ie threads can have mutex functions that are called by other threads.
2356Hence, all monitor features are available when using threads.
2357Figure~\ref{f:DirectCommunication} shows a comparison of direct call communication in \CFA with direct channel communication in Go.
2358(Ada provides a similar mechanism to the \CFA direct communication.)
2359The program main in both programs communicates directly with the other thread versus indirect communication where two threads interact through a passive monitor.
2360Both direct and indirection thread communication are valuable tools in structuring concurrent programs.
2361
2362\begin{figure}
2363\centering
2364\begin{lrbox}{\myboxA}
2365\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2366
2367struct Msg { int i, j; };
2368thread Gortn { int i;  float f;  Msg m; };
2369void mem1( Gortn & mutex gortn, int i ) { gortn.i = i; }
2370void mem2( Gortn & mutex gortn, float f ) { gortn.f = f; }
2371void mem3( Gortn & mutex gortn, Msg m ) { gortn.m = m; }
2372void ^?{}( Gortn & mutex ) {}
2373
2374void main( Gortn & gortn ) with( gortn ) {  // thread starts
2375
2376        for () {
2377
2378                `waitfor( mem1, gortn )` sout | i;  // wait for calls
2379                or `waitfor( mem2, gortn )` sout | f;
2380                or `waitfor( mem3, gortn )` sout | m.i | m.j;
2381                or `waitfor( ^?{}, gortn )` break;
2382
2383        }
2384
2385}
2386int main() {
2387        Gortn gortn; $\C[2.0in]{// start thread}$
2388        `mem1( gortn, 0 );` $\C{// different calls}\CRT$
2389        `mem2( gortn, 2.5 );`
2390        `mem3( gortn, (Msg){1, 2} );`
2391
2392
2393} // wait for completion
2394\end{cfa}
2395\end{lrbox}
2396
2397\begin{lrbox}{\myboxB}
2398\begin{Go}[aboveskip=0pt,belowskip=0pt]
2399func main() {
2400        type Msg struct{ i, j int }
2401
2402        ch1 := make( chan int )
2403        ch2 := make( chan float32 )
2404        ch3 := make( chan Msg )
2405        hand := make( chan string )
2406        shake := make( chan string )
2407        gortn := func() { $\C[1.5in]{// thread starts}$
2408                var i int;  var f float32;  var m Msg
2409                L: for {
2410                        select { $\C{// wait for messages}$
2411                          case `i = <- ch1`: fmt.Println( i )
2412                          case `f = <- ch2`: fmt.Println( f )
2413                          case `m = <- ch3`: fmt.Println( m )
2414                          case `<- hand`: break L $\C{// sentinel}$
2415                        }
2416                }
2417                `shake <- "SHAKE"` $\C{// completion}$
2418        }
2419
2420        go gortn() $\C{// start thread}$
2421        `ch1 <- 0` $\C{// different messages}$
2422        `ch2 <- 2.5`
2423        `ch3 <- Msg{1, 2}`
2424        `hand <- "HAND"` $\C{// sentinel value}$
2425        `<- shake` $\C{// wait for completion}\CRT$
2426}
2427\end{Go}
2428\end{lrbox}
2429
2430\subfloat[\CFA]{\label{f:CFAwaitfor}\usebox\myboxA}
2431\hspace{3pt}
2432\vrule
2433\hspace{3pt}
2434\subfloat[Go]{\label{f:Gochannel}\usebox\myboxB}
2435\caption{Direct communication}
2436\label{f:DirectCommunication}
2437\end{figure}
2438
2439\begin{comment}
2440The following shows an example of two threads directly calling each other and accepting calls from each other in a cycle.
2441\begin{cfa}
2442\end{cfa}
2443\vspace{-0.8\baselineskip}
2444\begin{cquote}
2445\begin{tabular}{@{}l@{\hspace{3\parindentlnth}}l@{}}
2446\begin{cfa}
2447thread Ping {} pi;
2448void ping( Ping & mutex ) {}
2449void main( Ping & pi ) {
2450        for ( 10 ) {
2451                `waitfor( ping, pi );`
2452                `pong( po );`
2453        }
2454}
2455int main() {}
2456\end{cfa}
2457&
2458\begin{cfa}
2459thread Pong {} po;
2460void pong( Pong & mutex ) {}
2461void main( Pong & po ) {
2462        for ( 10 ) {
2463                `ping( pi );`
2464                `waitfor( pong, po );`
2465        }
2466}
2467
2468\end{cfa}
2469\end{tabular}
2470\end{cquote}
2471% \lstMakeShortInline@%
2472% \caption{Threads ping/pong using external scheduling}
2473% \label{f:pingpong}
2474% \end{figure}
2475Note, the ping/pong threads are globally declared, @pi@/@po@, and hence, start (and possibly complete) before the program main starts.
2476\end{comment}
2477
2478
2479\subsection{Execution Properties}
2480
2481Table~\ref{t:ObjectPropertyComposition} shows how the \CFA high-level constructs cover 3 fundamental execution properties: thread, stateful function, and mutual exclusion.
2482Case 1 is a basic object, with none of the new execution properties.
2483Case 2 allows @mutex@ calls to Case 1 to protect shared data.
2484Case 3 allows stateful functions to suspend/resume but restricts operations because the state is stackless.
2485Case 4 allows @mutex@ calls to Case 3 to protect shared data.
2486Cases 5 and 6 are the same as 3 and 4 without restriction because the state is stackful.
2487Cases 7 and 8 are rejected because a thread cannot execute without a stackful state in a preemptive environment when context switching from the signal handler.
2488Cases 9 and 10 have a stackful thread without and with @mutex@ calls.
2489For situations where threads do not require direct communication, case 9 provides faster creation/destruction by eliminating @mutex@ setup.
2490
2491\begin{table}[b]
2492\caption{Object property composition}
2493\centering
2494\label{t:ObjectPropertyComposition}
2495\renewcommand{\arraystretch}{1.25}
2496%\setlength{\tabcolsep}{5pt}
2497\begin{tabular}{c|c|l|l}
2498\multicolumn{2}{c|}{object properties} & \multicolumn{2}{c}{mutual exclusion} \\
2499\hline
2500thread  & stateful                              & \multicolumn{1}{c|}{No} & \multicolumn{1}{c}{Yes} \\
2501\hline
2502\hline
2503No              & No                                    & \textbf{1}\ \ \ aggregate type                & \textbf{2}\ \ \ @monitor@ aggregate type \\
2504\hline
2505No              & Yes (stackless)               & \textbf{3}\ \ \ @generator@                   & \textbf{4}\ \ \ @monitor@ generator \\
2506\hline
2507No              & Yes (stackful)                & \textbf{5}\ \ \ @coroutine@                   & \textbf{6}\ \ \ @monitor@ @coroutine@ \\
2508\hline
2509Yes             & No / Yes (stackless)  & \textbf{7}\ \ \ {\color{red}rejected} & \textbf{8}\ \ \ {\color{red}rejected} \\
2510\hline
2511Yes             & Yes (stackful)                & \textbf{9}\ \ \ @thread@                              & \textbf{10}\ \ @monitor@ @thread@ \\
2512\end{tabular}
2513\end{table}
2514
2515
2516\subsection{Low-level Locks}
2517
2518For 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.
2519However, we strongly advocate using high-level concurrency mechanisms whenever possible.
2520
2521
2522\section{Parallelism}
2523\label{s:Parallelism}
2524
2525Historically, computer performance was about processor speeds.
2526However, with heat dissipation being a direct consequence of speed increase, parallelism is the new source for increased performance~\cite{Sutter05, Sutter05b}.
2527Therefore, high-performance applications must care about parallelism, which requires concurrency.
2528The lowest-level approach of parallelism is to use \newterm{kernel threads} in combination with semantics like @fork@, @join@, \etc.
2529However, kernel threads are better as an implementation tool because of complexity and higher cost.
2530Therefore, different abstractions are often layered onto kernel threads to simplify them, \eg pthreads.
2531
2532
2533\subsection{User Threads}
2534
2535A direct improvement on kernel threads is user threads, \eg Erlang~\cite{Erlang} and \uC~\cite{uC++book}.
2536This 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.
2537In many cases, user threads can be used on a much larger scale (100,000 threads).
2538Like kernel threads, user threads support preemption, which maximizes nondeterminism, but increases the potential for concurrency errors: race, livelock, starvation, and deadlock.
2539\CFA adopts user-threads as they represent the truest realization of concurrency and can build any other concurrency approach, \eg thread pools and actors~\cite{Actors}.
2540
2541A variant of user thread is \newterm{fibres}, which removes preemption, \eg Go~\cite{Go} @goroutine@s.
2542Like functional programming, which removes mutation and its associated problems, removing preemption from concurrency reduces nondeterminism, making race and deadlock errors more difficult to generate.
2543However, preemption is necessary for concurrency that relies on spinning, so there are a class of problems that cannot be programmed with fibres.
2544
2545
2546\subsection{Thread Pools}
2547
2548In contrast to direct threading is indirect \newterm{thread pools}, \eg Java @executor@, where small jobs (work units) are inserted into a work pool for execution.
2549If the jobs are dependent, \ie interact, there is an implicit/explicit dependency graph that ties them together.
2550While removing direct concurrency, and hence the amount of context switching, thread pools significantly limit the interaction that can occur among jobs.
2551Indeed, jobs should not block because that also blocks the underlying thread, which effectively means the CPU utilization, and therefore throughput, suffers.
2552While it is possible to tune the thread pool with sufficient threads, it becomes difficult to obtain high throughput and good core utilization as job interaction increases.
2553As well, concurrency errors return, which threads pools are suppose to mitigate.
2554
2555
2556\section{\protect\CFA Runtime Structure}
2557
2558Figure~\ref{f:RunTimeStructure} illustrates the runtime structure of a \CFA program.
2559In 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.
2560An executing thread is illustrated by its containment in a processor.
2561
2562\begin{figure}
2563\centering
2564\input{RunTimeStructure}
2565\caption{\CFA Runtime structure}
2566\label{f:RunTimeStructure}
2567\end{figure}
2568
2569
2570\subsection{Cluster}
2571\label{s:RuntimeStructureCluster}
2572
2573A \newterm{cluster} is a collection of threads and virtual processors (abstract kernel-thread) that execute the threads (like a virtual machine).
2574The purpose of a cluster is to control the amount of parallelism that is possible among threads, plus scheduling and other execution defaults.
2575The default cluster-scheduler is single-queue multi-server, which provides automatic load-balancing of threads on processors.
2576However, the scheduler is pluggable, supporting alternative schedulers.
2577If several clusters exist, both threads and virtual processors, can be explicitly migrated from one cluster to another.
2578No automatic load balancing among clusters is performed by \CFA.
2579
2580When 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.
2581The user cluster is created to contain the application user-threads.
2582Having all threads execute on the one cluster often maximizes utilization of processors, which minimizes runtime.
2583However, because of limitations of the underlying operating system, heterogeneous hardware, or scheduling requirements (real-time), multiple clusters are sometimes necessary.
2584
2585
2586\subsection{Virtual Processor}
2587\label{s:RuntimeStructureProcessor}
2588
2589A virtual processor is implemented by a kernel thread (\eg UNIX process), which is subsequently scheduled for execution on a hardware processor by the underlying operating system.
2590Programs may use more virtual processors than hardware processors.
2591On a multiprocessor, kernel threads are distributed across the hardware processors resulting in virtual processors executing in parallel.
2592(It is possible to use affinity to lock a virtual processor onto a particular hardware processor~\cite{affinityLinux, affinityWindows, affinityFreebsd, affinityNetbsd, affinityMacosx}, which is used when caching issues occur or for heterogeneous hardware processors.)
2593The \CFA runtime attempts to block unused processors and unblock processors as the system load increases;
2594balancing the workload with processors is difficult.
2595Preemption occurs on virtual processors rather than user threads, via operating-system interrupts.
2596Thus virtual processors execute user threads, where preemption frequency applies to a virtual processor, so preemption occurs randomly across the executed user threads.
2597Turning off preemption transforms user threads into fibres.
2598
2599
2600\subsection{Debug Kernel}
2601
2602There are two versions of the \CFA runtime kernel: debug and non-debug.
2603The debugging version has many runtime checks and internal assertions, \eg stack (non-writable) guard page, and checks for stack overflow whenever context switches occur among coroutines and threads, which catches most stack overflows.
2604After a program is debugged, the non-debugging version can be used to decrease space and increase performance.
2605
2606
2607\section{Implementation}
2608\label{s:Implementation}
2609
2610A primary implementation challenge is avoiding contention from dynamically allocating memory because of bulk acquire, \eg the internal-scheduling design is (almost) free of allocations.
2611All 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.
2612Furthermore, several bulk-acquire operations need a variable amount of memory.
2613This 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.
2614
2615In \CFA, ordering of monitor acquisition relies on memory ordering to prevent deadlock~\cite{Havender68}, because all objects have distinct non-overlapping memory layouts, and mutual-exclusion for a monitor is only defined for its lifetime.
2616When a mutex call is made, pointers to the concerned monitors are aggregated into a variable-length array and sorted.
2617This array persists for the entire duration of the mutual exclusion and is used extensively for synchronization operations.
2618
2619To 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;
2620the corresponding registers are then restored for the other context.
2621Note, the instruction pointer is untouched since the context switch is always inside the same function.
2622Experimental 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.
2623
2624All kernel threads (@pthreads@) created a stack.
2625Each \CFA virtual processor is implemented as a coroutine and these coroutines run directly on the kernel-thread stack, effectively stealing this stack.
2626The exception to this rule is the program main, \ie the initial kernel thread that is given to any program.
2627In 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.
2628
2629Finally, an important aspect for a complete threading system is preemption, which introduces extra non-determinism via transparent interleaving, rather than cooperation among threads for proper scheduling and processor fairness from long-running threads.
2630Because preemption frequency is usually long (1 millisecond) performance cost is negligible.
2631Preemption is normally handled by setting a count-down timer on each virtual processor.
2632When the timer expires, an interrupt is delivered, and the interrupt handler resets the count-down 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.
2633Multiple signal handlers may be pending.
2634When 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.
2635The 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;
2636therefore, the same signal mask is required for all virtual processors in a cluster.
2637
2638However, on current UNIX systems:
2639\begin{cquote}
2640A process-directed signal may be delivered to any one of the threads that does not currently have the signal blocked.
2641If more than one of the threads has the signal unblocked, then the kernel chooses an arbitrary thread to which to deliver the signal.
2642SIGNAL(7) - Linux Programmer's Manual
2643\end{cquote}
2644Hence, the timer-expiry signal, which is generated \emph{externally} by the UNIX kernel to the UNIX process, is delivered to any of its UNIX subprocesses (kernel threads).
2645To ensure each virtual processor receives its own preemption signals, a discrete-event simulation is run on a special virtual processor, and only it sets and receives timer events.
2646Virtual processors register an expiration time with the discrete-event simulator, which is inserted in sorted order.
2647The simulation sets the count-down 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.
2648Processing a preemption event sends an \emph{internal} @SIGUSR1@ signal to the registered virtual processor, which is always delivered to that processor.
2649
2650
2651\section{Performance}
2652\label{results}
2653
2654To verify the implementation of the \CFA runtime, a series of microbenchmarks are performed comparing \CFA with Java OpenJDK-9, Go 1.9.2 and \uC 7.0.0.
2655The benchmark computer is an AMD Opteron\texttrademark\ 6380 NUMA 64-core, 8 socket, 2.5 GHz processor, running Ubuntu 16.04.3 LTS and \uC and \CFA are compiled with gcc 6.3.
2656
2657\begin{comment}
2658\begin{table}
2659\centering
2660\caption{Experiment environment}
2661\label{t:machine}
2662
2663\begin{tabular}{|l|r||l|r|}
2664\hline
2665Architecture            & x86\_64                               & NUMA node(s)  & 8 \\
2666\hline
2667CPU op-mode(s)          & 32-bit, 64-bit                & Model name    & AMD Opteron\texttrademark\ Processor 6380 \\
2668\hline
2669Byte Order                      & Little Endian                 & CPU Freq              & 2.5 GHz \\
2670\hline
2671CPU(s)                          & 64                                    & L1d cache     & 16 KiB \\
2672\hline
2673Thread(s) per core      & 2                                     & L1i cache     & 64 KiB \\
2674\hline
2675Core(s) per socket      & 8                                     & L2 cache              & 2048 KiB \\
2676\hline
2677Socket(s)                       & 4                                     & L3 cache              & 6144 KiB \\
2678\hline
2679\hline
2680Operating system        & Ubuntu 16.04.3 LTS    & Kernel                & Linux 4.4-97-generic \\
2681\hline
2682gcc                                     & 6.3                                   & \CFA                  & 1.0.0 \\
2683\hline
2684Java                            & OpenJDK-9                     & Go                    & 1.9.2 \\
2685\hline
2686\end{tabular}
2687\end{table}
2688\end{comment}
2689
2690All benchmarks are run using the following harness.
2691\newpage
2692\begin{cfa}
2693unsigned int N = 10_000_000;
2694#define BENCH( `run` ) Time before = getTimeNsec();  `run;`  Duration result = (getTimeNsec() - before) / N;
2695\end{cfa}
2696The method used to get time is @clock_gettime( CLOCK_REALTIME )@.
2697Each benchmark is performed @N@ times, where @N@ varies depending on the benchmark;
2698the total time is divided by @N@ to obtain the average time for a benchmark.
2699All omitted tests for other languages are functionally identical to the shown \CFA test.
2700
2701
2702\paragraph{Context-Switching}
2703
2704In procedural programming, the cost of a function call is important as modularization (refactoring) increases.
2705(In many cases, a compiler inlines function calls to eliminate this cost.)
2706Similarly, when modularization extends to coroutines/tasks, the time for a context switch becomes a relevant factor.
2707The coroutine test is from resumer to suspender and from suspender to resumer, which is two context switches.
2708The thread test is using yield to enter and return from the runtime kernel, which is two context switches.
2709The difference in performance between coroutine and thread context-switch is the cost of scheduling for threads, whereas coroutines are self-scheduling.
2710Figure~\ref{f:ctx-switch} only shows the \CFA code for coroutines/threads (other systems are similar) with all results in Table~\ref{tab:ctx-switch}.
2711
2712\begin{multicols}{2}
2713\lstset{language=CFA,moredelim=**[is][\color{red}]{@}{@},deletedelim=**[is][]{`}{`}}
2714\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2715coroutine C {} c;
2716void main( C & ) { for ( ;; ) { @suspend();@ } }
2717int main() { // coroutine test
2718        BENCH( for ( N ) { @resume( c );@ } )
2719        sout | result`ns;
2720}
2721int main() { // task test
2722        BENCH( for ( N ) { @yield();@ } )
2723        sout | result`ns;
2724}
2725\end{cfa}
2726\captionof{figure}{\CFA context-switch benchmark}
2727\label{f:ctx-switch}
2728
2729\columnbreak
2730
2731\vspace*{-16pt}
2732\captionof{table}{Context switch comparison (nanoseconds)}
2733\label{tab:ctx-switch}
2734\begin{tabular}{@{}r*{3}{D{.}{.}{3.2}}@{}}
2735\multicolumn{1}{@{}c}{} & \multicolumn{1}{c}{Median} &\multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
2736Kernel Thread   & 333.5 & 332.96        & 4.1   \\
2737\CFA Coroutine  & 49    & 48.68         & 0.47  \\
2738\CFA Thread             & 105   & 105.57        & 1.37  \\
2739\uC Coroutine   & 44    & 44            & 0             \\
2740\uC Thread              & 100   & 99.29         & 0.96  \\
2741Goroutine               & 145   & 147.25        & 4.15  \\
2742Java Thread             & 373.5 & 375.14        & 8.72
2743\end{tabular}
2744\end{multicols}
2745
2746
2747\paragraph{Mutual-Exclusion}
2748
2749Mutual exclusion is measured by entering/leaving a critical section.
2750For monitors, entering and leaving a monitor function is measured.
2751To put the results in context, the cost of entering a non-inline function and the cost of acquiring and releasing a @pthread_mutex@ lock is also measured.
2752Figure~\ref{f:mutex} shows the code for \CFA with all results in Table~\ref{tab:mutex}.
2753Note, the incremental cost of bulk acquire for \CFA, which is largely a fixed cost for small numbers of mutex objects.
2754
2755\begin{multicols}{2}
2756\lstset{language=CFA,moredelim=**[is][\color{red}]{@}{@},deletedelim=**[is][]{`}{`}}
2757\begin{cfa}
2758monitor M {} m1/*, m2, m3, m4*/;
2759void __attribute__((noinline))
2760do_call( M & mutex m/*, m2, m3, m4*/ ) {}
2761int main() {
2762        BENCH(
2763                for( N ) @do_call( m1/*, m2, m3, m4*/ );@
2764        )
2765        sout | result`ns;
2766}
2767\end{cfa}
2768\captionof{figure}{\CFA acquire/release mutex benchmark}
2769\label{f:mutex}
2770
2771\columnbreak
2772
2773\vspace*{-16pt}
2774\captionof{table}{Mutex comparison (nanoseconds)}
2775\label{tab:mutex}
2776\begin{tabular}{@{}r*{3}{D{.}{.}{3.2}}@{}}
2777\multicolumn{1}{@{}c}{} & \multicolumn{1}{c}{Median} &\multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
2778C function                                              & 2                     & 2             & 0             \\
2779FetchAdd + FetchSub                             & 26            & 26    & 0             \\
2780Pthreads Mutex Lock                             & 31            & 31.71 & 0.97  \\
2781\uC @monitor@ member rtn.               & 31            & 31    & 0             \\
2782\CFA @mutex@ function, 1 arg.   & 46            & 46.68 & 0.93  \\
2783\CFA @mutex@ function, 2 arg.   & 84            & 85.36 & 1.99  \\
2784\CFA @mutex@ function, 4 arg.   & 158           & 161   & 4.22  \\
2785Java synchronized function              & 27.5          & 29.79 & 2.93
2786\end{tabular}
2787\end{multicols}
2788
2789
2790\paragraph{Internal Scheduling}
2791
2792Internal scheduling is measured by waiting on and signalling a condition variable.
2793Figure~\ref{f:int-sched} shows the code for \CFA, with results in Table~\ref{tab:int-sched}.
2794Note, the incremental cost of bulk acquire for \CFA, which is largely a fixed cost for small numbers of mutex objects.
2795Java scheduling is significantly greater because the benchmark explicitly creates multiple thread in order to prevent the JIT from making the program sequential, \ie removing all locking.
2796
2797\begin{multicols}{2}
2798\lstset{language=CFA,moredelim=**[is][\color{red}]{@}{@},deletedelim=**[is][]{`}{`}}
2799\begin{cfa}
2800volatile int go = 0;
2801monitor M { condition c; } m;
2802void __attribute__((noinline))
2803do_call( M & mutex a1 ) { signal( c ); }
2804thread T {};
2805void main( T & this ) {
2806        while ( go == 0 ) { yield(); }
2807        while ( go == 1 ) { @do_call( m );@ }
2808}
2809int  __attribute__((noinline))
2810do_wait( M & mutex m ) with(m) {
2811        go = 1; // continue other thread
2812        BENCH( for ( N ) { @wait( c );@ } );
2813        go = 0; // stop other thread
2814        sout | result`ns;
2815}
2816int main() {
2817        T t;
2818        do_wait( m );
2819}
2820\end{cfa}
2821\captionof{figure}{\CFA Internal-scheduling benchmark}
2822\label{f:int-sched}
2823
2824\columnbreak
2825
2826\vspace*{-16pt}
2827\captionof{table}{Internal-scheduling comparison (nanoseconds)}
2828\label{tab:int-sched}
2829\bigskip
2830
2831\begin{tabular}{@{}r*{3}{D{.}{.}{5.2}}@{}}
2832\multicolumn{1}{@{}c}{} & \multicolumn{1}{c}{Median} & \multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
2833Pthreads Cond. Variable         & 6005          & 5681.43       & 835.45        \\
2834\uC @signal@                            & 324           & 325.54        & 3,02          \\
2835\CFA @signal@, 1 @monitor@      & 368.5         & 370.61        & 4.77          \\
2836\CFA @signal@, 2 @monitor@      & 467           & 470.5         & 6.79          \\
2837\CFA @signal@, 4 @monitor@      & 700.5         & 702.46        & 7.23          \\
2838Java @notify@                           & 15471         & 172511        & 5689
2839\end{tabular}
2840\end{multicols}
2841
2842
2843\paragraph{External Scheduling}
2844
2845External scheduling is measured by accepting a call using the @waitfor@ statement (@_Accept@ in \uC).
2846Figure~\ref{f:ext-sched} shows the code for \CFA, with results in Table~\ref{tab:ext-sched}.
2847Note, the incremental cost of bulk acquire for \CFA, which is largely a fixed cost for small numbers of mutex objects.
2848
2849\begin{multicols}{2}
2850\lstset{language=CFA,moredelim=**[is][\color{red}]{@}{@},deletedelim=**[is][]{`}{`}}
2851\vspace*{-16pt}
2852\begin{cfa}
2853volatile int go = 0;
2854monitor M {} m;
2855thread T {};
2856void __attribute__((noinline))
2857do_call( M & mutex ) {}
2858void main( T & ) {
2859        while ( go == 0 ) { yield(); }
2860        while ( go == 1 ) { @do_call( m );@ }
2861}
2862int __attribute__((noinline))
2863do_wait( M & mutex m ) {
2864        go = 1; // continue other thread
2865        BENCH( for ( N ) { @waitfor( do_call, m );@ } )
2866        go = 0; // stop other thread
2867        sout | result`ns;
2868}
2869int main() {
2870        T t;
2871        do_wait( m );
2872}
2873\end{cfa}
2874\captionof{figure}{\CFA external-scheduling benchmark}
2875\label{f:ext-sched}
2876
2877\columnbreak
2878
2879\vspace*{-16pt}
2880\captionof{table}{External-scheduling comparison (nanoseconds)}
2881\label{tab:ext-sched}
2882\begin{tabular}{@{}r*{3}{D{.}{.}{3.2}}@{}}
2883\multicolumn{1}{@{}c}{} & \multicolumn{1}{c}{Median} &\multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
2884\uC @_Accept@                           & 358           & 359.11        & 2.53          \\
2885\CFA @waitfor@, 1 @monitor@     & 359           & 360.93        & 4.07          \\
2886\CFA @waitfor@, 2 @monitor@     & 450           & 449.39        & 6.62          \\
2887\CFA @waitfor@, 4 @monitor@     & 652           & 655.64        & 7.73
2888\end{tabular}
2889\end{multicols}
2890
2891
2892
2893\paragraph{Object Creation}
2894
2895Object creation is measured by creating/deleting the specific kind of concurrent object.
2896Figure~\ref{f:creation} shows the code for \CFA, with results in Table~\ref{tab:creation}.
2897The only note here is that the call stacks of \CFA coroutines are lazily created, therefore without priming the coroutine to force stack creation, the creation cost is artificially low.
2898
2899\begin{multicols}{2}
2900\lstset{language=CFA,moredelim=**[is][\color{red}]{@}{@},deletedelim=**[is][]{`}{`}}
2901\begin{cfa}
2902thread MyThread {};
2903void main( MyThread & ) {}
2904int main() {
2905        BENCH( for ( N ) { @MyThread m;@ } )
2906        sout | result`ns;
2907}
2908\end{cfa}
2909\captionof{figure}{\CFA object-creation benchmark}
2910\label{f:creation}
2911
2912\columnbreak
2913
2914\vspace*{-16pt}
2915\captionof{table}{Object creation comparison (nanoseconds)}
2916\label{tab:creation}
2917
2918\begin{tabular}[t]{@{}r*{3}{D{.}{.}{5.2}}@{}}
2919\multicolumn{1}{@{}c}{} & \multicolumn{1}{c}{Median} & \multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
2920Pthreads                                & 28091         & 28073.39      & 163.1         \\
2921\CFA Coroutine Lazy             & 6                     & 6.07          & 0.26          \\
2922\CFA Coroutine Eager    & 520           & 520.61        & 2.04          \\
2923\CFA Thread                             & 2032          & 2016.29       & 112.07        \\
2924\uC Coroutine                   & 106           & 107.36        & 1.47          \\
2925\uC Thread                              & 536.5         & 537.07        & 4.64          \\
2926Goroutine                               & 3103          & 3086.29       & 90.25         \\
2927Java Thread                             & 103416.5      & 103732.29     & 1137          \\
2928\end{tabular}
2929\end{multicols}
2930
2931
2932\section{Conclusion}
2933
2934Advanced control-flow will always be difficult, especially when there is temporal ordering and nondeterminism.
2935However, many systems exacerbate the difficulty through their presentation mechanisms.
2936This paper shows it is possible to present a hierarchy of control-flow features, generator, coroutine, thread, and monitor, providing an integrated set of high-level, efficient, and maintainable control-flow features.
2937Eliminated from \CFA are spurious wakeup and barging, which are nonintuitive and lead to errors, and having to work with a bewildering set of low-level locks and acquisition techniques.
2938\CFA high-level race-free monitors and tasks provide the core mechanisms for mutual exclusion and synchronization, without having to resort to magic qualifiers like @volatile@/@atomic@.
2939Extending these mechanisms to handle high-level deadlock-free bulk acquire across both mutual exclusion and synchronization is a unique contribution.
2940The \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.
2941The M:N model is judged to be efficient and provide greater flexibility than a 1:1 threading model.
2942These concepts and the entire \CFA runtime-system are written in the \CFA language, extensively leveraging the \CFA type-system, which demonstrates the expressiveness of the \CFA language.
2943Performance comparisons with other concurrent systems/languages show the \CFA approach is competitive across all low-level operations, which translates directly into good performance in well-written concurrent applications.
2944C 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.
2945
2946
2947\section{Future Work}
2948
2949While control flow in \CFA has a strong start, development is still underway to complete a number of missing features.
2950
2951\paragraph{Flexible Scheduling}
2952\label{futur:sched}
2953
2954An important part of concurrency is scheduling.
2955Different scheduling algorithms can affect performance (both in terms of average and variation).
2956However, no single scheduler is optimal for all workloads and therefore there is value in being able to change the scheduler for given programs.
2957One solution is to offer various tuning options, allowing the scheduler to be adjusted to the requirements of the workload.
2958However, to be truly flexible, a pluggable scheduler is necessary.
2959Currently, 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.
2960
2961\paragraph{Non-Blocking I/O}
2962\label{futur:nbio}
2963
2964Many modern workloads are not bound by computation but IO operations, a common case being web servers and XaaS~\cite{XaaS} (anything as a service).
2965These types of workloads require significant engineering to amortizing costs of blocking IO-operations.
2966At its core, non-blocking I/O is an operating-system level feature queuing IO operations, \eg network operations, and registering for notifications instead of waiting for requests to complete.
2967Current 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.
2968However, these solutions lead to code that is hard to create, read and maintain.
2969A better approach is to tie non-blocking I/O into the concurrency system to provide ease of use with low overhead, \eg thread-per-connection web-services.
2970A non-blocking I/O library is currently under development for \CFA.
2971
2972\paragraph{Other Concurrency Tools}
2973\label{futur:tools}
2974
2975While monitors offer flexible and powerful concurrency for \CFA, other concurrency tools are also necessary for a complete multi-paradigm concurrency package.
2976Examples of such tools can include futures and promises~\cite{promises}, executors and actors.
2977These additional features are useful for applications that can be constructed without shared data and direct blocking.
2978As well, new \CFA extensions should make it possible to create a uniform interface for virtually all mutual exclusion, including monitors and low-level locks.
2979
2980\paragraph{Implicit Threading}
2981\label{futur:implcit}
2982
2983Basic concurrent (embarrassingly parallel) applications can benefit greatly from implicit concurrency, where sequential programs are converted to concurrent, possibly with some help from pragmas to guide the conversion.
2984This type of concurrency can be achieved both at the language level and at the library level.
2985The canonical example of implicit concurrency is concurrent nested @for@ loops, which are amenable to divide and conquer algorithms~\cite{uC++book}.
2986The \CFA language features should make it possible to develop a reasonable number of implicit concurrency mechanism to solve basic HPC data-concurrency problems.
2987However, implicit concurrency is a restrictive solution with significant limitations, so it can never replace explicit concurrent programming.
2988
2989
2990\section{Acknowledgements}
2991
2992The authors would like to recognize the design assistance of Aaron Moss, Rob Schluntz and Andrew Beach on the features described in this paper.
2993Funding for this project has been provided by Huawei Ltd.\ (\url{http://www.huawei.com}), and Peter Buhr is partially funded by the Natural Sciences and Engineering Research Council of Canada.
2994
2995{%
2996\fontsize{9bp}{12bp}\selectfont%
2997\bibliography{pl,local}
2998}%
2999
3000\end{document}
3001
3002% Local Variables: %
3003% tab-width: 4 %
3004% fill-column: 120 %
3005% compile-command: "make" %
3006% End: %
Note: See TracBrowser for help on using the repository browser.