%\usepackage{epsfig}
-\newcommand{\note}[1]{{\em Note: #1}}
+%\newcommand{\note}[1]{{\em Note: #1}}
+\newcommand{\note}[1]{{{\bf Note:}\sl #1}}
+\newcommand{\ToDo}[1]{{{\bf ToDo:}\sl #1}}
+\newcommand{\Arg}[1]{\mbox{${\tt arg}_{#1}$}}
+\newcommand{\bottom}{bottom} % foo, can't remember the symbol name
+
% DIMENSION OF TEXT:
\textheight 8.5 in
\textwidth 6.25 in
\begin{document}
-\newcommand{\ToDo}[1]{{{\bf ToDo:}\sl #1}}
-\newcommand{\Note}[1]{{{\bf Note:}\sl #1}}
-\newcommand{\Arg}[1]{\mbox{${\tt arg}_{#1}$}}
-\newcommand{\bottom}{bottom} % foo, can't remember the symbol name
-
\title{The STG runtime system (revised)}
\author{Simon Peyton Jones \\ Glasgow University and Oregon Graduate Institute \and
Simon Marlow \\ Glasgow University \and
that a program can consist of a mixture of GHC-compiled and Hugs-interpreted
code.
+\item The RTS supports concurrency by default.
+This has some costs (eg we can't do hardware stack checks) but
+reduces the number of different configurations we need to support.
+
\item CAFs are only retained if they are
reachable. Since they are referred to by implicit references buried
in code, this means that the garbage collector must traverse the whole
You can make the following choices:
\begin{itemize}
\item
-Support for concurrency or parallelism. There are four
-mutually-exclusive choices.
+Support for parallelism. There are three mutually-exclusive choices.
\begin{description}
-\item[@SEQUENTIAL@] No concurrency or parallelism support.
- This configuration might not support interrupt recovery.
-
- \note{There's probably not much point in supporting this option. If
- we've gone to the effort of supporting concurency, we don't gain
- much by being able to turn it off.}
-
-\item[@CONCURRENT@] Support for concurrency but not for parallelism.
-\item[@CONCURRENT@+@GRANSIM@] Concurrency support and simulated parallelism.
-\item[@CONCURRENT@+@PARALLEL@] Concurrency support and real parallelism.
+\item[@SEQUENTIAL@] Support for concurrency but not for parallelism.
+\item[@GRANSIM@] Concurrency support and simulated parallelism.
+\item[@PARALLEL@] Concurrency support and real parallelism.
\end{description}
\item @PROFILING@ adds cost-centre profiling.
If you find something which disagrees with this terminology, fix the
usage.}
-\begin{itemize}
-
-\item A {\em word} is (at least) 32 bits and can hold either a signed
-or an unsigned int.
-
-\item A {\em pointer} is (at least) 32 bits and big enough to hold a
-function pointer or a data pointer.
-
-\item A {\em boxed} type is one whose elements are heap allocated.
+In the type system, we have boxed and unboxed types.
-\item An {\em unboxed} type is one whose elements are {\em not} heap allocated.
+\begin{itemize}
-\item A {\em pointed} type is one that contains $\bot$. Variables with
+\item A \emph{pointed} type is one that contains $\bot$. Variables with
pointed types are the only things which can be lazily evaluated. In
the STG machine, this means that they are the only things that can be
-{\em entered} or {\em updated} and it requires that they be boxed.
+\emph{entered} or \emph{updated} and it requires that they be boxed.
-\item An {\em unpointed} type is one that does not contain $\bot$.
+\item An \emph{unpointed} type is one that does not contain $\bot$.
Variables with unpointed types are never delayed --- they are always
evaluated when they are constructed. In the STG machine, this means
-that they cannot be {\em entered} or {\em updated}. Unpointed objects
+that they cannot be \emph{entered} or \emph{updated}. Unpointed objects
may be boxed (like @Array#@) or unboxed (like @Int#@).
-\item A {\em closure} is a (representation of) a value of a {\em pointed}
-type. It may be in HNF or it may be unevaluated --- in either case, you can
-try to evaluate it again.
+\end{itemize}
+
+In the implementation, we have different kinds of objects:
+
+\begin{itemize}
+
+\item \emph{boxed} objects are heap objects used by the evaluators
+
+\item \emph{unboxed} objects are not heap allocated
+
+\item \emph{stack} objects are allocated on the stack
+
+\item \emph{closures} are objects which can be \emph{entered}.
+They are always boxed and always have boxed types.
+They may be in WHNF or they may be unevaluated.
+
+\item A \emph{thunk} is a (representation of) a value of a \emph{pointed}
+type which is \emph{not} in WHNF.
+
+\item A \emph{value} is an object in WHNF. It can be pointed or unpointed.
+
+\end{itemize}
+
+
+
+At the hardware level, we have \emph{word}s and \emph{pointer}s.
+
+\begin{itemize}
-\item A {\em thunk} is a (representation of) a value of a {\em pointed}
-type which is {\em not} in HNF.
+\item A \emph{word} is (at least) 32 bits and can hold either a signed
+or an unsigned int.
-\item A {\em value} is an object in HNF. It can be pointed or unpointed.
+\item A \emph{pointer} is (at least) 32 bits and big enough to hold a
+function pointer or a data pointer.
\end{itemize}
Occasionally, a field of a data structure must hold either a word or a
-pointer. In such circumstances, it is {\em not safe} to assume that
+pointer. In such circumstances, it is \emph{not safe} to assume that
words and pointers are the same size.
+
+
% Todo:
% More terminology to mention.
% unboxed, unpointed
\begin{itemize}
-\item The garbage collector never expands an object when it promotes
-it to the old generation. This is important because the GC avoids
-performing heap overflow checks by assuming that the amount added to
-the old generation is no bigger than the current new generation.
-
\item
If the garbage collector is allowed to shrink the stack of a thread,
> cons = \ x xs -> (:) x xs
@
+\note{For historical reasons, GHC doesn't use this syntax --- but it should.}
\subsection{Unboxed tuples}\label{sect:unboxed-tuples}
to the same restrictions as other unpointed types.
Operationally, unboxed tuples are never built on the heap. When
-unboxed tuples are returned, they are returned in multiple registers
+an unboxed tuple is returned, it is returned in multiple registers
or multiple stack slots. At first sight, this seems a little strange
but it's no different from passing double precision floats in two
registers.
\subsection{STG Syntax}
+
\ToDo{Insert STG syntax with appropriate changes.}
The major components of the system are:
\begin{itemize}
-\item The scheduler
-\item The storage manager
-\item The evaluators
-\item The loader
-\item The compilers
+
+\item
+
+The evaluators (section~\ref{sect:sm-overview}) are responsible for
+evaluating heap objects. The system supports two evaluators: the
+machine code evaluator; and the bytecode evaluator.
+
+\item
+
+The scheduler (section~\ref{sect:scheduler-overview}) acts as the
+coordinator for the whole system. It is responsible for switching
+between evaluators, switching between threads, garbage collection,
+communication between multiple processors, etc.
+
+\item
+
+The storage manager (section~\ref{sect:evaluators-overview}) is
+responsible for allocating blocks of contiguous memory and for garbage
+collection.
+
+\item
+
+The loader (section~\ref{sect:loader-overview}) is responsible for
+loading machine code and bytecode files from the file system and for
+resolving references between separately compiled modules.
+
+\item
+
+The compilers (section~\ref{sect:compilers-overview}) generate machine
+code and bytecode files which can be loaded by the loader.
+
\end{itemize}
\ToDo{Insert diagram showing all components underneath the scheduler
and communicating only with the scheduler}
-\section{Scheduler}
+
+\section{The Evaluators}\label{sect:evaluators-overview}
+
+There are two evaluators: a machine code evaluator and a bytecode
+evaluator. The evaluators task is to evaluate code within a thread
+until one of the following happens:
+
+\begin{itemize}
+\item heap overflow
+\item stack overflow
+\item it is preempted
+\item it blocks in one of the concurrency primitives
+\item it performs a safe ccall
+\item it needs to switch to the other evaluator.
+\end{itemize}
+
+The evaluators expect to find a closure on top of the thread's stack
+and terminate with a closure on top of the thread's stack.
+
+\subsection{Evaluation Model}
+\label{sect:evaluation-model}
+
+Whilst the evaluators differ internally, they share a common
+evaluation model and many object representations.
+
+\subsubsection{Heap Objects}
+
+The choice of heap and stack objects used by the evaluators is tightly
+bound to the evaluation model. This section provides an overview of
+the most important heap and stack objects; further details are given
+later.
+
+All heap objects look like this:
+
+\begin{center}
+\begin{tabular}{|l|l|l|l|}\hline
+\emph{Header} & \emph{Payload} \\ \hline
+\end{tabular}
+\end{center}
+
+The header's vary between different kinds of object but they all start
+with a pointer to a pair consisting of an \emph{info table} and some
+\emph{entry code}. The info table is used both by the evaluators and
+by the storage manager and contains an @INFO_TYPE@ field which
+identifies which kind of heap object uses it and determines the
+interpretation of the payload and of the other fields of the info
+table. The entry code is some machine code used by the machine code
+evaluator to evaluate closures and raises an error for other kinds of
+objects.
+
+The major kinds of heap object used are as follows. (For simplicity,
+this description omits certain optimisations and extra fields required
+by the garbage collector.)
+
+\begin{description}
+
+\item[Constructors] are used to represent data constructors. Their
+payload consists of the fields of the constructor; the tag of the
+constructor is stored in the info table.
+
+\begin{center}
+\begin{tabular}{|l|l|l|l|}\hline
+@CONSTR@ & \emph{Fields} \\ \hline
+\end{tabular}
+\end{center}
+
+\item[Primitive objects] are used to represent objects with unpointed
+types which are too large to fit in a register (or stack slot) or for
+which sharing must be preserved. Primitive objects include large
+objects such as multiple precision integers and immutable arrays and
+mutable objects such as mutable arrays, mutable variables, MVar's,
+IVar's and foreign object pointers. Since unpointed objects are not
+pointed, they cannot be entered. Their payload varies according to
+the kind of object.
+
+\item[Function closures] are used to represent functions. Their
+payload (if any) consists of the free variables of the function.
+
+\begin{center}
+\begin{tabular}{|l|l|l|l|}\hline
+@FUN@ & \emph{Free Variables} \\ \hline
+\end{tabular}
+\end{center}
+
+Function closures are only generated by the machine code compiler.
+
+\item[Thunks] are used to represent unevaluated expressions which will
+be updated with their result. Their payload (if any) consists of the
+free variables of the function. The entry code for a thunk starts by
+pushing an \emph{update frame} onto the stack and overwriting the
+thunk with a \emph{black hole}. When evaluation of the thunk
+completes, the update frame will cause the thunk to be overwritten
+again with an \emph{indirection} to the result of the thunk.
+
+\begin{center}
+\begin{tabular}{|l|l|l|l|}\hline
+@THUNK@ & \emph{Free Variables} \\ \hline
+\end{tabular}
+\end{center}
+
+Thunks are only generated by the machine code compiler.
+
+\item[Byte-code Objects (@BCO@s)] are generated by the bytecode
+compiler. In conjunction with \emph{updateable applications} and
+\emph{non-updateeable applications} they are used to represent
+functions, unevaluated expressions and return addresses.
+
+\begin{center}
+\begin{tabular}{|l|l|l|l|}\hline
+@BCO@ & \emph{Constant Pool} & \emph{Bytecodes} \\ \hline
+\end{tabular}
+\end{center}
+
+\item[Non-updatable Applications] are used to represent the
+application of a function to an insufficient number of arguments.
+Their payload consists of the function and the arguments received so far.
+
+\begin{center}
+\begin{tabular}{|l|l|l|l|}\hline
+@PAP@ & \emph{Function Closure} & \emph{Arguments} \\ \hline
+\end{tabular}
+\end{center}
+
+@PAP@s are used when a function is applied to too few arguments and by
+code generated by the lambda-lifting phase of the bytecode compiler.
+
+\item[Updatable Applications] are used to represent the application of
+a function to a sufficient number of arguments. Their payload
+consists of the function and its arguments.
+
+Updateable applications are like thunks: on entering an updateable
+application, the evaluators push an \emph{update frame} onto the stack
+and overwrite the application with a \emph{black hole}; when
+evaluation completes, the evaluators overwrite the application with an
+\emph{indirection} to the result of the application.
+
+\begin{center}
+\begin{tabular}{|l|l|l|l|}\hline
+@AP@ & \emph{Function Closure} & \emph{Arguments} \\ \hline
+\end{tabular}
+\end{center}
+
+@AP@s are only generated by the bytecode compiler.
+
+\item[Black holes] are used to mark updateable closures which are
+currently being evaluated. ``Black holing'' an object cures a
+potential space leak and detects certain classes of infinite loops.
+More imporantly, black holes act as synchronisation objects between
+separate threads: if a second thread tries to enter an updateable
+closure which is already being evaluated, the second thread is added
+to a list of blocked threads and the thread is suspended.
+
+When evaluation of the black-holed closure completes, the black hole
+is overwritten with an indirection to the result of the closure and
+any blocked threads are restored to the runnable queue.
+
+\begin{center}
+\begin{tabular}{|l|l|l|l|}\hline
+@BH@ & \emph{Blocked threads} \\ \hline
+\end{tabular}
+\end{center}
+
+\ToDo{In a single threaded system, it's trivial to detect infinite
+loops: reentering a BH is always an error. How easy is it in a
+multi-threaded system?}
+
+\item[Indirections] are used to update an unevaluated closure with its
+(usually fully evaluated) result in situations where it isn't possible
+to perform an update in place. (In the current system, we always
+update with an indirection to avoid duplicating the result when doing
+an update in place.)
+
+\begin{center}
+\begin{tabular}{|l|l|l|l|}\hline
+@IND@ & \emph{Closure} \\ \hline
+\end{tabular}
+\end{center}
+
+Indirections needn't always point to an evaluated closure. They can
+point to a chain of indirections which point to an evaluated closure.
+When revertible black holes are added, they may also point to reverted
+black holes.
+
+\item[Thread State Objects (@TSO@s)] represent Haskell threads. Their
+payload consists of a unique thread id, the status of the thread
+(runnable, blocked, etc) and the stack. @TSO@s may be resized by the
+scheduler if its stack is too small or too large.
+
+\begin{center}
+\begin{tabular}{|l|l|l|l|}\hline
+@TSO@ & \emph{Thread Id} & \emph{Status} & \emph{Stack} \\ \hline
+\end{tabular}
+\end{center}
+
+\end{description}
+
+\subsubsection{Stack Objects}
+
+The stack contains a mixture of \emph{pending arguments} and
+\emph{stack objects}.
+
+Pending arguments are arguments to curried functions which have not
+yet been incorporated into an activation frame. For example, when
+evaluating @let { g x y = x + y; f x = g{x} } in f{3,4}@, the
+evaluator pushes both arguments onto the stack and enters @f@. @f@
+only requires one argument so it leaves the second argument as a
+\emph{pending argument}. The pending argument remains on the stack
+until @f@ calls @g@ which requires two arguments: the argument passed
+to it by @f@ and the pending argument which was passed to @f@.
+
+Unboxed pending arguments are always preceeded by a ``tag'' which says
+how large the argument is. This allows the garbage collector to
+locate pointers within the stack.
+
+There are three kinds of stack object: return addresses, update frames
+and seq frames. All stack objects look like this
+
+\begin{center}
+\begin{tabular}{|l|l|l|l|}\hline
+\emph{Header} & \emph{Payload} \\ \hline
+\end{tabular}
+\end{center}
+
+As with heap objects, the header starts with a pointer to a pair
+consisting of an \emph{info table} and some \emph{entry code}.
+
+\begin{description}
+
+\item[Return addresses] are used to cause selection and execution of
+case alternatives when a constructor is returned. Return addresses
+generated by the machine code compiler look like this:
+
+\begin{center}
+\begin{tabular}{|l|l|l|l|}\hline
+\emph{@RET_ADDR@} & \emph{Free Variables of the case alternatives} \\ \hline
+\end{tabular}
+\end{center}
+
+The free variables are a mixture of pointers and non-pointers whose
+layout is described by the info table.
+
+Return addresses generated by the bytecode compiler look like this:
+\begin{center}
+\begin{tabular}{|l|l|l|l|}\hline
+\emph{@BCO_RET@} & \emph{BCO} & \emph{Free Variables of the case alternatives} \\ \hline
+\end{tabular}
+\end{center}
+
+There is just one @BCO_RET@ info pointer. We avoid needing different
+@BCO_RET@s for each stack layout by tagging unboxed free variables as
+though they were pending arguments.
+
+\item[Update frames] are used to trigger updates. When an update
+frame is entered, it overwrites the updatee with an indirection to the
+result, restarts any threads blocked on the @BH@ and returns to the
+stack object underneath the update frame.
+
+\begin{center}
+\begin{tabular}{|l|l|l|l|}\hline
+\emph{@UPDATE@} & \emph{Next Update Frame} & \emph{Updatee} \\ \hline
+\end{tabular}
+\end{center}
+
+\item[Seq frames] are used to implement the polymorphic @seq@ primitive.
+They are a special kind of update frame.
+
+\ToDo{Describe them properly}
+
+
+\end{description}
+
+\ToDo{We also need a stop frame which goes on the bottom of the stack
+when the thread terminates.}
+
+
+\subsubsection{Case expressions}
+
+In the STG language, all evaluation is triggered by evaluating a case
+expression. When evaluating a case expression @case e of alts@, the
+evaluator push a return address onto the stack and evaluate the
+expression @e@. When @e@ eventually reduces to a constructor, the
+return address on the stack is entered. The details of how the
+constructor is passed to the return address and how the appropriate
+case alternative is selected vary between evaluators.
+
+Case expressions for unboxed data types are essentially the same: the
+case expression pushes a return address onto the stack before
+evaluating the scrutinee; when a function returns an unboxed value, it
+enters the return address on top of the stack.
+
+
+\subsubsection{Function Applications}
+
+In the STG language, all function calls are tail calls. The arguments
+are pushed onto the stack and the function closure is entered. If any
+arguments are unboxed, they must be tagged as unboxed pending
+arguments. Entering a closure is just a special case of calling a
+function with no arguments.
+
+
+\subsubsection{Let expressions}
+
+In the STG language, almost all heap allocation is caused by let
+expressions. Filling in the contents of a set of mutually recursive
+heap objects is simple enough; the only difficulty is that once the heap space has been allocated, the thread must not return to the scheduler until
+after the objects are filled in.
+
+
+\subsubsection{Primitive Operations}
+
+\ToDo{}
+
+Most primops are simple, some aren't.
+
+
+
+
+
+
+\section{Scheduler}\label{sect:scheduler-overview}
The Scheduler is the heart of the run-time system. A running program
consists of a single running thread, and a list of runnable and
-blocked threads. All threads consist of a stack and a few words of
+blocked threads. A thread is represented by a \emph{Thread Status Object} (TSO), which contains a few words consist of a stack and a few words of
status information. Except for the running thread, all threads have a
closure on top of their stack; the scheduler restarts a thread by
-entering an evaluator which performs some reduction and returns.
+entering an evaluator which performs some reduction and returns to the
+scheduler.
\subsection{The scheduler's main loop}
When a C program calls some Haskell code.
+\item
+
+By @forkIO@, @takeMVar@ and (maybe) other Concurrent Haskell primitives.
+
\end{itemize}
\subsection{Restarting a thread}
-The evaluators can reduce almost all types of closure except that only
-the machine code evaluator can reduce GHC-compiled closures and only
-the bytecode evaluator can reduce Hugs-compiled closures.
-Consequently, the scheduler may use either evaluator to restart a
-thread unless the top closure is a @BCO@ or contains machine code.
-
-However, if the top of the stack contains a constructor, the scheduler
-should use the machine code evaluator to restart the thread. This
-allows the bytecode evaluator to return a constructor to a machine
-code return address by pushing the constructor on top of the stack and
-returning to the scheduler. If the return address under the
+When the scheduler decides to run a thread, it has to decide which
+evaluator to use. It does this by looking at the type of the closure
+on top of the stack.
+\begin{itemize}
+\item @BCO@ $\Rightarrow$ bytecode evaluator
+\item @FUN@ or @THUNK@ $\Rightarrow$ machine code evaluator
+\item @CONSTR@ $\Rightarrow$ machine code evaluator
+\item other $\Rightarrow$ either evaluator.
+\end{itemize}
+
+The only surprise in the above is that the scheduler must enter the
+machine code evaluator if there's a constructor on top of the stack.
+This allows the bytecode evaluator to return a constructor to a
+machine code return address by pushing the constructor on top of the
+stack and returning to the scheduler. If the return address under the
constructor is @HUGS_RET@, the entry code for @HUGS_RET@ will
rearrange the stack so that the return @BCO@ is on top of the stack
and return to the scheduler which will then call the bytecode
evaluator. There is little point in trying to shorten this slightly
-indirect route since it will happen very rarely if at all.
+indirect route since it is will happen very rarely if at all.
+
+\note{As an optimisation, we could store the choice of evaluator in
+the TSO status whenever we leave the evaluator. This is required for
+any thread, no matter what state it is in (blocked, stack overflow,
+etc). It isn't clear whether this would accomplish anything.}
\subsection{Returning from a thread}
\begin{itemize}
\item A heap check fails, and a garbage collection is required
\item Compiled code needs to switch to interpreted code, and vice versa.
-\item The evaluator needs to perform an ``unsafe'' C call.
+\item The evaluator needs to perform a ``safe'' C call.
\item The thread becomes blocked.
\item The thread is preempted.
\item The thread terminates.
Except when the thread terminates, the thread always terminates with a
closure on the top of the stack.
-\subsection{Preempting a thread}
-
-Strictly speaking, threads cannot be preempted --- the scheduler
-merely sets a preemption request flag which the thread must arrange to
-test on a regular basis. When an evaluator finds that the preemption
-request flag is set, it pushes an appropriate closure onto the stack
-and returns to the scheduler.
-
-In the bytecode interpreter, the flag is tested whenever we enter a
-closure. If the preemption flag is set, it leaves the closure on top
-of the stack and returns to the scheduler.
-
-In the machine code evaluator, the flag is only tested when a heap or
-stack check fails. This is less expensive than testing the flag on
-entering every closure but runs the risk that a thread will enter an
-infinite loop which does not allocate any space. If the flag is set,
-the evaluator returns to the scheduler exactly as if a heap check had
-failed.
-
-\subsection{``Safe'' and ``unsafe'' C calls}
-
-There are two ways of calling C:
-
-\begin{description}
-
-\item[``Safe'' C calls]
-are used if the programer is certain that the C function will not
-do anything dangerous such as calling a Haskell function or an
-operating system call which blocks the thread for a long period of time.
-\footnote{Warning: this use of ``safe'' and ``unsafe'' is the exact
-opposite of the usage for functions like @unsafePerformIO@.}
-Safe C calls are faster but must be hand-checked by the programmer.
-
-Safe C calls are performed by pushing the arguments onto the C stack
-and jumping to the C function's entry point. On exit, the result of
-the function is in a register which is returned to the Haskell code as
-an unboxed value.
-
-\item[``Unsafe'' C calls] are used if the programmer suspects that the
-thread may do something dangerous like blocking or calling a Haskell
-function. Unsafe C calls are relatively slow but are less problematic.
-
-Unsafe C calls are performed by pushing the arguments onto the Haskell
-stack, pushing a return continuation and returning a \emph{C function
-descriptor} to the scheduler. The scheduler suspends the Haskell thread,
-spawns a new operating system thread which pops the arguments off the
-Haskell stack onto the C stack, calls the C function, pushes the
-function result onto the Haskell stack and informs the scheduler that
-the C function has completed and the Haskell thread is now runnable.
-
-\end{description}
-
-The bytecode evaluator will probably treat all C calls as being unsafe.
-
-\ToDo{It might be good for the programmer to indicate how the program is
-unsafe. For example, if we distinguish between C functions which might
-call Haskell functions and those which might block, we could perform a
-safe call for blocking functions in a single-threaded system or, perhaps, in a multi-threaded system which only happens to have a single thread at the moment.}
-
-
-\section{The Evaluators}
-
-All the scheduler needs to know about evaluation is how to manipulate
-threads and how to find the closure on top of the stack. The
-evaluators need to agree on the representations of certain objects and
-on how to return from the scheduler.
-
\subsection{Returning to the Scheduler}
\label{sect:switching-worlds}
+\ToDo{This ignores the other three ways of returning}
+
The evaluators return to the scheduler under three circumstances:
\begin{itemize}
\note{Hugs doesn't support unboxed values in source programs but they
are used for a few complex primops.}
-When it enters a constructor, the bytecode evaluator tests the return
-continuation on top of the stack. If it is a machine code
-continuation, it returns to the scheduler with the unboxed value and a
-special closure on top of the stack. When the closure is entered (by
-the machine code evaluator), it returns the unboxed value on top of
-the stack to the return continuation under it.
+When it returns an unboxed value, the bytecode evaluator tests the
+return continuation on top of the stack. If it is a machine code
+continuation, it returns to the scheduler with the tagged unboxed
+value and a special closure on top of the stack. When the closure is
+entered (by the machine code evaluator), it returns the unboxed value
+on top of the stack to the return continuation under it.
-The runtime system (or GHC?) provides one of these closures for each
-unboxed type. Hugs cannot generate them itself since the entry code is
-really very tricky.
+The runtime library for GHC provides one of these closures for each unboxed
+type. Hugs cannot generate them itself since the entry code is really
+very tricky.
\paragraph{Heap/Stack overflow and preemption}
We avoid the need to test return addresses in the machine code
evaluator by pushing a special return address on top of a pointer to
the bytecode return continuation. This return address rearranges the
-stack so that the bco pointer is above the unboxed value (as shown in
+stack so that the bco pointer is above the tagged unboxed value (as shown in
figure~\ref{fig:hugs-entering-unboxed-return}) and returns to the scheduler.
\begin{figure}[ht]
\ToDo{}
-\subsection{Shared Representations}
-We share @AP@s, @PAP@s, constructors, indirections, selectors(?) and
-update frames. These are described in section~\ref{sect:heap-objects}.
+\subsection{Preempting a thread}
+Strictly speaking, threads cannot be preempted --- the scheduler
+merely sets a preemption request flag which the thread must arrange to
+test on a regular basis. When an evaluator finds that the preemption
+request flag is set, it pushes an appropriate closure onto the stack
+and returns to the scheduler.
-\section{The Storage Manager}
+In the bytecode interpreter, the flag is tested whenever we enter a
+closure. If the preemption flag is set, it leaves the closure on top
+of the stack and returns to the scheduler.
-The storage manager is responsible for managing the heap and all
-objects stored in it. Most objects are just copied in the normal way
-but a number receive special treatment by the storage manager:
+In the machine code evaluator, the flag is only tested when a heap or
+stack check fails. This is less expensive than testing the flag on
+entering every closure but runs the risk that a thread will enter an
+infinite loop which does not allocate any space. If the flag is set,
+the evaluator returns to the scheduler exactly as if a heap check had
+failed.
+
+\subsection{``Safe'' and ``unsafe'' C calls}
+
+There are two ways of calling C:
+
+\begin{description}
+
+\item[``Unsafe'' C calls] are used if the programer is certain that
+the C function will not do anything dangerous. Unsafe C calls are
+faster but must be hand-checked by the programmer.
+
+Dangerous things include:
\begin{itemize}
-\item
-Indirections are shorted out.
+\item
+
+Call a system function such as @getchar@ which might block
+indefinitely. This is dangerous because we don't want the entire
+runtime system to block just because one thread blocks.
\item
-Weak pointers and stable pointers are treated specially.
+Call an RTS function which will block on the RTS access semaphore.
+This would lead to deadlock.
\item
-Thread State Objects (TSOs) and the stacks within them are treated specially.
-In particular:
+Call a Haskell function. This is just a special case of calling an
+RTS function.
+
+\end{itemize}
+
+Unsafe C calls are performed by pushing the arguments onto the C stack
+and jumping to the C function's entry point. On exit, the result of
+the function is in a register which is returned to the Haskell code as
+an unboxed value.
+
+\item[``Safe'' C calls] are used if the programmer suspects that the
+thread may do something dangerous. Safe C calls are relatively slow
+but are less problematic.
+
+Safe C calls are performed by pushing the arguments onto the Haskell
+stack, pushing a return continuation and returning a \emph{C function
+descriptor} to the scheduler. The scheduler suspends the Haskell thread,
+spawns a new operating system thread which pops the arguments off the
+Haskell stack onto the C stack, calls the C function, pushes the
+function result onto the Haskell stack and informs the scheduler that
+the C function has completed and the Haskell thread is now runnable.
+
+\end{description}
+
+The bytecode evaluator will probably treat all C calls as being safe.
+
+\ToDo{It might be good for the programmer to indicate how the program
+is unsafe. For example, if we distinguish between C functions which
+might call Haskell functions and those which might block, we could
+perform an unsafe call for blocking functions in a single-threaded
+system or, perhaps, in a multi-threaded system which only happens to
+have a single thread at the moment.}
+
+
+
+\section{The Storage Manager}\label{sect:sm-overview}
+
+The storage manager is responsible for managing the heap and all
+objects stored in it. It provides special support for lazy evaluation
+and for foreign function calls.
+
+\subsection{SM support for lazy evaluation}
\begin{itemize}
\item
+Indirections are shorted out.
+
+\item
+
Update frames pointing to unreachable objects are squeezed out.
\item
Adjacent update frames (for different closures) are compressed to a
single update frame pointing to a single black hole.
+\end{itemize}
+
+
+\subsection{SM support for foreign function calls}
+
+\begin{itemize}
+
+\item
+
+Stable pointers allow other languages to access Haskell objects.
+
+\item
+
+Foreign Objects are a form of weak pointer which let's Haskell access
+foreign objects.
+
+\end{itemize}
+
+\subsection{Misc}
+
+\begin{itemize}
+
\item
If the stack contains a large amount of free space, the storage
\ToDo{Would it be useful for the storage manager to enlarge the stack?}
-\end{itemize}
-
\item
-Very large objects (eg large arrays and TSOs) are not moved if
-possible.
+For efficiency reasons, very large objects (eg large arrays and TSOs)
+are not moved if possible.
\end{itemize}
-\section{The Compilers}
+\section{The Compilers}\label{sect:compilers-overview}
Need to describe interface files, format of bytecode files, symbols
defined by machine code files.
(Again, all that matters is what the loader sees.)
-\section{The Loader}
-
-\ToDo{Is it ok to load code when threads are running?}
+\section{The Loader}\label{sect:loader-overview}
In a batch mode system, we can statically link all the modules
together. In an interactive system we need a loader which will
indirection is possible but tricky.
Note: We could avoid patching machine code if all references to
-eternal references went through the SRT --- then we just have one
+external references went through the SRT --- then we just have one
thing to patch. But the SRT always contains a pointer to the closure
rather than the fast entry point (say), so we'd take a big performance
hit for doing this.
\end{description}
+Using the above scheme, all accesses to ``external'' objects involve a
+layer of indirection. To avoid this overhead, the machine code
+compiler might provide a way for the programmer to specify which
+modules will be statically linked and which will be dynamically linked
+--- the idea being that statically linked code and data will be
+accessed directly.
+
+
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\part{Internal details}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
The major components of the system are:
\begin{itemize}
-\item The scheduler
-\item The storage manager
+\item The scheduler (section~\ref{sect:storage-manager-internals})
+\item The storage manager (section~\ref{sect:storage-manager-internals})
\item The evaluators
\item The loader
\item The compilers
\end{itemize}
\section{The Scheduler}
+\label{sect:scheduler-internals}
+
+\ToDo{Detailed description of scheduler}
+
+Many heap objects contain fields allowing them to be inserted onto lists
+during evaluation or during garbage collection. The lists required by
+the evaluator and storage manager are as follows.
+
+\begin{itemize}
+
+\item 4 lists of threads: runnable threads, sleeping threads, threads
+waiting for timeout and threads waiting for I/O.
+
+\item The \emph{mutables list} is a list of all objects in the old
+generation which might contain pointers into the new generation. Most
+of the objects on this list are indirections (section~\ref{sect:IND})
+or ``mutable.'' (Section~\ref{sect:mutables}.)
+
+\item The \emph{Foreign Object list} is a list of all foreign objects
+ which have not yet been deallocated. (Section~\ref{sect:FOREIGN}.)
+
+\item The \emph{Spark pool} is a doubly(?) linked list of Spark objects
+maintained by the parallel system. (Section~\ref{sect:SPARK}.)
+
+\item The \emph{Blocked Fetch list} (or
+lists?). (Section~\ref{sect:BLOCKED_FETCH}.)
+
+\item For each thread, there is a list of all update frames on the
+stack. (Section~\ref{sect:data-updates}.)
+
+\item The Stable Pointer Table is a table of pointers to objects which
+are known to the outside world and must be retained by the garbage
+collector even if they are not accessible from within the heap.
+
+\end{itemize}
+
+\ToDo{The links for these fields are usually inserted immediately
+after the fixed header except ...}
+
+
\section{The Storage Manager}
\label{sect:storage-manager-internals}
-\ToDo{Fix this picture}
+\subsection{Misc Text looking for a home}
+
+A \emph{value} may be:
+\begin{itemize}
+\item \emph{Boxed}, i.e.~represented indirectly by a pointer to a heap object (e.g.~foreign objects, arrays); or
+\item \emph{Unboxed}, i.e.~represented directly by a bit-pattern in one or more registers (e.g.~@Int#@ and @Float#@).
+\end{itemize}
+All \emph{pointed} values are \emph{boxed}.
+
+
+\subsection{Heap Objects}
\begin{figure}
\begin{center}
\input{closure}
\end{center}
+\ToDo{Fix this picture}
\caption{A closure}
\label{fig:closure}
\end{figure}
-Every {\em heap object} is a contiguous block
-of memory, consisting of a fixed-format {\em header} followed
-by zero or more {\em data words}.
-
-\ToDo{I absolutely do not believe that every heap object has a header
-like this - ADR. I believe that they all have an info pointer but I
-see no readon why stack objects and unpointed heap objects would have
-an entry count since this will always be zero.}
+Every \emph{heap object} is a contiguous block
+of memory, consisting of a fixed-format \emph{header} followed
+by zero or more \emph{data words}.
The header consists of the following fields:
\begin{itemize}
-\item A one-word {\em info pointer}, which points to
-the object's static {\em info table}.
-\item Zero or more {\em admin words} that support
+\item A one-word \emph{info pointer}, which points to
+the object's static \emph{info table}.
+\item Zero or more \emph{admin words} that support
\begin{itemize}
-\item Profiling (notably a {\em cost centre} word).
+\item Profiling (notably a \emph{cost centre} word).
\note{We could possibly omit the cost centre word from some
administrative objects.}
\item Parallelism (e.g. GranSim keeps the object's global address here,
used to indicate if a closure has been updated but not yet entered. It
is set when the closure is updated and cleared when subsequently
entered.
-
-NB: It is {\em not} an ``entry count'', it is an
+\footnote{%
+NB: It is \emph{not} an ``entry count'', it is an
``entries-after-update count.'' The commoning up of @CONST@,
@CHARLIKE@ and @INTLIKE@ closures is turned off(?) if this is
required. This has only been done for 2s collection.
+}
\end{itemize}
\end{itemize}
Most of the RTS is completely insensitive to the number of admin words.
The total size of the fixed header is @FIXED_HS@.
-Many heap objects contain fields allowing them to be inserted onto lists
-during evaluation or during garbage collection. The lists required by
-the evaluator and storage manager are as follows.
-
-\begin{itemize}
-\item 2 lists of threads: runnable threads and sleeping threads.
-
-\item The {\em static object list} is a list of all statically
-allocated objects which might contain pointers into the heap.
-(Section~\ref{sect:static-objects}.)
-
-\item The {\em updated thunk list} is a list of all thunks in the old
-generation which have been updated with an indirection.
-(Section~\ref{sect:IND_OLDGEN}.)
-
-\item The {\em mutables list} is a list of all other objects in the
-old generation which might contain pointers into the new generation.
-Most of the object on this list are ``mutable.''
-(Section~\ref{sect:mutables}.)
-
-\item The {\em Foreign Object list} is a list of all foreign objects
- which have not yet been deallocated. (Section~\ref{sect:FOREIGN}.)
-
-\item The {\em Spark pool} is a doubly(?) linked list of Spark objects
-maintained by the parallel system. (Section~\ref{sect:SPARK}.)
-
-\item The {\em Blocked Fetch list} (or
-lists?). (Section~\ref{sect:BLOCKED_FETCH}.)
-
-\item For each thread, there is a list of all update frames on the
-stack. (Section~\ref{sect:data-updates}.)
-
-
-\end{itemize}
-
-\ToDo{The links for these fields are usually inserted immediately
-after the fixed header except ...}
-
\subsection{Info Tables}
-An {\em info table} is a contiguous block of memory, {\em laid out
+An \emph{info table} is a contiguous block of memory, \emph{laid out
backwards}. That is, the first field in the list that follows
occupies the highest memory address, and the successive fields occupy
successive decreasing memory addresses.
An info table has the following contents (working backwards in memory
addresses):
\begin{itemize}
-\item The {\em entry code} for the closure.
+\item The \emph{entry code} for the closure.
This code appears literally as the (large) last entry in the
info table, immediately preceded by the rest of the info table.
-An {\em info pointer} always points to the first byte of the entry code.
+An \emph{info pointer} always points to the first byte of the entry code.
-\item A one-word {\em closure type field}, @INFO_TYPE@, identifies what kind
+\item A one-word \emph{closure type field}, @INFO_TYPE@, identifies what kind
of closure the object is. The various types of closure are described
in Section~\ref{sect:closures}.
In some configurations, some useful properties of
closures (is it a HNF? can it be sparked?)
are represented as high-order bits so they can be tested quickly.
-\item A single pointer or word --- the {\em storage manager info field},
+\item A single pointer or word --- the \emph{storage manager info field},
@INFO_SM@, contains auxiliary information describing the closure's
precise layout, for the benefit of the garbage collector and the code
that stuffs graph into packets for transmission over the network.
-\item A one-word {\em Tag/Static Reference Table} field, @INFO_SRT@.
+\item A one-word \emph{Tag/Static Reference Table} field, @INFO_SRT@.
For data constructors, this field contains the constructor tag, in the
range $0..n-1$ where $n$ is the number of constructors. For all other
objects it contains a pointer to a table which enables the garbage
collector to identify all accessible code and CAFs. They are fully
described in Section~\ref{sect:srt}.
-\item {\em Profiling info\/}
+\item \emph{Profiling info\/}
\ToDo{The profiling info is completely bogus. I've not deleted it
from the document but I've commented it all out.}
\fi % end of commented out stuff
-\item {\em Parallelism info\/}
+\item \emph{Parallelism info\/}
\ToDo{}
-\item {\em Debugging info\/}
+\item \emph{Debugging info\/}
\ToDo{}
\end{itemize}
this:
\begin{itemize}
\item
-{\em Static closures} occupy fixed, statically-allocated memory
+\emph{Static closures} occupy fixed, statically-allocated memory
locations, with globally known addresses.
\item
-{\em Dynamic closures} are individually allocated in the heap.
+\emph{Dynamic closures} are individually allocated in the heap.
\item
-{\em Stack closures} are closures allocated within a thread's stack
+\emph{Stack closures} are closures allocated within a thread's stack
(which is itself a heap object). Unlike other closures, there are
never any pointers to stack closures. Stack closures are discussed in
Section~\ref{sect:stacks}.
A second useful classification is this:
\begin{itemize}
\item
-{\em Executive objects}, such as thunks and data constructors,
+\emph{Executive objects}, such as thunks and data constructors,
participate directly in a program's execution. They can be subdivided into
three kinds of objects according to their type:
\begin{itemize}
\item
-{\em Pointed objects}, represent values of a {\em pointed} type
+\emph{Pointed objects}, represent values of a \emph{pointed} type
(<.pointed types launchbury.>) --i.e.~a type that includes $\bottom$ such as @Int@ or @Int# -> Int#@.
-\item {\em Unpointed objects}, represent values of a {\em unpointed} type --i.e.~a type that does not include $\bottom$ such as @Int#@ or @Array#@.
+\item \emph{Unpointed objects}, represent values of a \emph{unpointed} type --i.e.~a type that does not include $\bottom$ such as @Int#@ or @Array#@.
-\item {\em Activation frames}, represent ``continuations''. They are
+\item \emph{Activation frames}, represent ``continuations''. They are
always stored on the stack and are never pointed to by heap objects or
passed as arguments. \note{It's not clear if this will still be true
once we support speculative evaluation.}
\end{itemize}
-\item {\em Administrative objects}, such as stack objects and thread
+\item \emph{Administrative objects}, such as stack objects and thread
state objects, do not represent values in the original program.
\end{itemize}
closure kind & Section \\
\hline
-{\em Pointed} \\
+\emph{Pointed} \\
\hline
@CONSTR@ & \ref{sect:CONSTR} \\
@IND_STATIC@ & \ref{sect:IND} \\
\hline
-{\em Unpointed} \\
+\emph{Unpointed} \\
\hline
@ARR_WORDS@ & \ref{sect:ARR_WORDS1},\ref{sect:ARR_WORDS2} \\
\item @isWHNF@ is true if the object is in Weak Head Normal Form.
Note that unpointed objects are (arbitrarily) not considered to be in WHNF.
-@isWHNF@ is true for @PAP@s, @CONSTR@s, @FUN@s and some @BCO@s.
+@isWHNF@ is true for @PAP@s, @CONSTR@s, @FUN@s and all @BCO@s.
\ToDo{Need to distinguish between whnf BCOs and non-whnf BCOs in their
@INFO_TYPE@}
-\item @isBOTTOM@ is true if the object is known to be $\bot$. It is
-true of @BH@s. \note{I suspect we'll want to add other kinds of
-infotype which are known to be bottom later.}
-
\item @isUPDATEABLE@ is true if the object may be overwritten with an
indirection object.
#define _ST 0x0004 /* Is static */
#define _MU 0x0008 /* Is mutable */
#define _UP 0x0010 /* Is updatable (but not mutable) */
-#define _BM 0x0020 /* Is a "primitive" array */
+#define _BM 0x0020 /* Is a "rimitive" array */
#define _BH 0x0040 /* Is a black hole */
#define _IN 0x0080 /* Is an indirection */
#define _TH 0x0100 /* Is a thunk */
\fi
-\subsection{Pointed Objects}
+\subsection{Closures (aka Pointed Objects)}
-All pointed objects can be entered.
+An object can be entered iff it is a closure.
\subsubsection{Function closures}\label{sect:FUN}
in g x
@
Both @f@ and @g@ are represented by function closures. The closure
-for @f@ is {\em static} while that for @g@ is {\em dynamic}.
+for @f@ is \emph{static} while that for @g@ is \emph{dynamic}.
The layout of a function closure is as follows:
\begin{center}
\begin{tabular}{|l|l|l|l|}\hline
-{\em Fixed header} & {\em Pointers} & {\em Non-pointers} \\ \hline
+\emph{Fixed header} & \emph{Pointers} & \emph{Non-pointers} \\ \hline
\end{tabular}
\end{center}
The data words (pointers and non-pointers) are the free variables of
layout than dynamic ones:
\begin{center}
\begin{tabular}{|l|l|l|}\hline
-{\em Fixed header} & {\em Static object link} \\ \hline
+\emph{Fixed header} & \emph{Static object link} \\ \hline
\end{tabular}
\end{center}
Static function closures have no free variables. (However they may refer to other
static closures; these references are recorded in the function closure's SRT.)
-They have one field that is not present in dynamic closures, the {\em static object
+They have one field that is not present in dynamic closures, the \emph{static object
link} field. This is used by the garbage collector in the same way that to-space
is, to gather closures that have been determined to be live but that have not yet
been scavenged.
closures. That is
\begin{center}
\begin{tabular}{|l|l|l|l|}\hline
-{\em Fixed header} & {\em Pointers} & {\em Non-pointers} \\ \hline
+\emph{Fixed header} & \emph{Pointers} & \emph{Non-pointers} \\ \hline
\end{tabular}
\end{center}
So its layout is like this:
\begin{center}
\begin{tabular}{|l|l|l|l|l|}\hline
-{\em Fixed header} & {\em Pointers} & {\em Non-pointers} & {\em Static object link}\\ \hline
+\emph{Fixed header} & \emph{Pointers} & \emph{Non-pointers} & \emph{Static object link}\\ \hline
\end{tabular}
\end{center}
The static object link, at the end of the closure, serves the same purpose
link field. Since we expect that there might be quite a lot of static
constructors this optimisation makes sense. Furthermore, the @NOCAF@
tag allows the compiler to indicate that no CAFs can be reached
-anywhere {\em even indirectly}.
+anywhere \emph{even indirectly}.
\end{itemize}
Here the right-hand sides of @range@ and @ys@ are both thunks; the former
is static while the latter is dynamic.
-The layout of a thunk is the same as that for a function closure,
-except that it may have some words of ``slop'' at the end to make sure
-that it has
-at least @MIN_UPD_PAYLOAD@ words in addition to its
-fixed header.
+The layout of a thunk is the same as that for a function closure.
+However, thunks must have a payload of at least @MIN_UPD_PAYLOAD@ words
+to allow it to be overwritten with a black hole and an indirection.
+The compiler may have to add extra non-pointer fields to satisfy this constraint.
\begin{center}
\begin{tabular}{|l|l|l|l|l|}\hline
-{\em Fixed header} & {\em Pointers} & {\em Non-pointers} & {\em Slop} \\ \hline
+\emph{Fixed header} & \emph{Pointers} & \emph{Non-pointers} \\ \hline
\end{tabular}
\end{center}
The @INFO_SM@ word contains the same information as for function
-closures; that is, number of pointers and number of non-pointers (excluding slop).
+closures; that is, number of pointers and number of non-pointers.
A thunk differs from a function closure in that it can be updated.
There are several forms of thunk:
\begin{itemize}
-\item @THUNK@: a vanilla, dynamically allocated thunk.
-The garbage collection code for thunks whose
-pointer + non-pointer words is less than @MIN_UPD_PAYLOAD@ differs from
-that for function closures and data constructors, because the GC code
-has to account for the slop.
-\item $@THUNK_@p@_@np$. Similar comments apply.
+\item @THUNK@ and $@THUNK_@p@_@np$: vanilla, dynamically allocated thunks.
+Dynamic thunks are overwritten with normal indirections.
+
\item @THUNK_STATIC@. A static thunk is also known as
-a {\em constant applicative form}, or {\em CAF}.
+a \emph{constant applicative form}, or \emph{CAF}.
+Static thunks are overwritten with static indirections.
\begin{center}
\begin{tabular}{|l|l|l|l|l|}\hline
-{\em Fixed header} & {\em Pointers} & {\em Non-pointers} & {\em Slop} & {\em Static object link}\\ \hline
+\emph{Fixed header} & \emph{Pointers} & \emph{Non-pointers} \emph{Static object link}\\ \hline
\end{tabular}
\end{center}
is a selector thunk. A selector thunk is laid out like this:
\begin{center}
\begin{tabular}{|l|l|l|l|}\hline
-{\em Fixed header} & {\em Selectee pointer} \\ \hline
+\emph{Fixed header} & \emph{Selectee pointer} \\ \hline
\end{tabular}
\end{center}
The @INFO_SM@ word contains the byte offset of the desired word in
A Byte-Code Object (BCO) is a container for a a chunk of byte-code,
which can be executed by Hugs. The byte-code represents a
-supercombinator in the program: when hugs compiles a module, it
+supercombinator in the program: when Hugs compiles a module, it
performs lambda lifting and each resulting supercombinator becomes a
byte-code object in the heap.
-There are two kinds of BCO: a standard @BCO@ which has an arity of one
-or more, and a @BCO_CAF@ which takes no arguments and can be updated.
-When a @BCO_CAF@ is updated, the code is thrown away!
+BCOs are not updateable; the bytecode compiler represents updatable thunks
+using a combination of @AP@s and @BCO@s.
The semantics of BCOs are described in Section
\ref{sect:hugs-heap-objects}. A BCO has the following structure:
code.
\end{itemize}
+
\subsubsection{Partial applications (PAPs)}\label{sect:PAP}
+\ToDo{PAPs don't contains update frames or activation frames. When we
+add revertible black holes, we'll introduce a new kind of object which
+can contain activation frames.}
+
A partial application (PAP) represents a function applied to too few arguments.
It is only built as a result of updating after an argument-satisfaction
check failure. A PAP has the following shape:
\begin{center}
\begin{tabular}{|l|l|l|l|}\hline
-{\em Fixed header} & {\em No of arg words} & {\em Function closure} & {\em Arg stack} \\ \hline
+\emph{Fixed header} & \emph{No of arg words} & \emph{Function closure} & \emph{Arg stack} \\ \hline
\end{tabular}
\end{center}
The ``arg stack'' is a copy of the chunk of stack above the update
Its entry code simply copies the arg stack chunk back on top of the
stack and enters the function closure. (It has to do a stack overflow test first.)
-PAPs are also used to implement Hugs functions (where the arguments are free variables).
-PAPs generated by Hugs can be static.
+PAPs are also used to implement Hugs functions (where the arguments
+are free variables). PAPs generated by Hugs can be static so we need
+both @PAP@ and @PAP_STATIC@.
\subsubsection{@AP@ objects}
\label{sect:AP}
\begin{center}
\begin{tabular}{|l|l|l|l|}
\hline
-\emph{Fixed Header} & {\em No of arg words} & {\em Function closure} & {\em Arg stack} \\
+\emph{Fixed Header} & \emph{No of arg words} & \emph{Function closure} & \emph{Arg stack} \\
\hline
\end{tabular}
\end{center}
for the thunk. The argument stack may be empty if the thunk has no
free variables.
+\note{Since @AP@s are updateable, the @MIN_UPD_PAYLOAD@ constraint
+applies here too.}
\subsubsection{Indirections}
\label{sect:IND}
-\label{sect:IND_OLDGEN}
Indirection closures just point to other closures. They are introduced
when a thunk is updated to point to its value.
shape:
\begin{center}
\begin{tabular}{|l|l|l|}\hline
-{\em Fixed header} & {\em Target closure} \\ \hline
-\end{tabular}
-\end{center}
-
-\item[@IND_OLDGEN@] is the indirection used to update an old-generation
-thunk. Its shape is like this:
-\begin{center}
-\begin{tabular}{|l|l|l|}\hline
-{\em Fixed header} & {\em Mutable link field} & {\em Target closure} \\ \hline
+\emph{Fixed header} & \emph{Mutable link field} & \emph{Target closure} \\ \hline
\end{tabular}
\end{center}
-It contains a {\em mutable link field} that is used to string together
-all old-generation indirections that might have a pointer into
-the new generation.
+It contains a \emph{mutable link field} that is used to string together
+indirections in each generation.
-\item[@IND_PERMANENT@ and @IND_OLDGEN_PERMANENT@.]
+\item[@IND_PERMANENT@]
for lexical profiling, it is necessary to maintain cost centre
information in an indirection, so ``permanent indirections'' are
retained forever. Otherwise they are just like vanilla indirections.
\note{If a permanent indirection points to another permanent
indirection or a @CONST@ closure, it is possible to elide the indirection
since it will have no effect on the profiler.}
-\note{Do we still need @IND@ and @IND_OLDGEN@
-in the profiling build, or can we just make
-do with one pair whose behaviour changes when profiling is built?}
+
+\note{Do we still need @IND@ in the profiling build, or do we just
+need @IND@ but its behaviour changes when profiling is on?}
\item[@IND_STATIC@] is used for overwriting CAFs when they have been
evaluated. Static indirections are not removed by the garbage
\begin{center}
\begin{tabular}{|l|l|l|}
\hline
-{\em Fixed header} & {\em Target closure} & {\em Static object link} \\
+\emph{Fixed header} & \emph{Target closure} & \emph{Static object link} \\
\hline
\end{tabular}
\end{center}
\end{description}
-\subsubsection{Activation Records}
-
-Activation records are parts of the stack described by return address
-info tables (closures with @INFO_TYPE@ values of @RET_SMALL@,
-@RET_VEC_SMALL@, @RET_BIG@ and @RET_VEC_BIG@). They are described in
-section~\ref{sect:activation-records}.
-
-
-\subsubsection{Black holes, MVars and IVars}
+\subsubsection{Black holes and Blocking Queues}
\label{sect:BH}
-\label{sect:MVAR}
-\label{sect:IVAR}
Black hole closures are used to overwrite closures currently being
evaluated. They inform the garbage collector that there are no live
\begin{center}
\begin{tabular}{|l|l|l|}
\hline
-{\em Fixed header} & {\em Mutable link} & {\em Blocked thread link} \\
+\emph{Fixed header} & \emph{Mutable link} & \emph{Blocked thread link} \\
\hline
\end{tabular}
\end{center}
-The {\em Blocked thread link} points to the TSO of the first thread
+The \emph{Blocked thread link} points to the TSO of the first thread
waiting for the value of this thunk. All subsequent TSOs in the list
are linked together using their @TSO_LINK@ field.
-When the blocking queue is non-@NULL@, the black hole must be added to
-the mutables list since the TSOs on the list may contain pointers into
-the new generation. There is no need to clutter up the mutables list
-with black holes with empty blocking queues.
+When the blocking queue is non-@NULL@ and the @BH@ is in the old
+generation, the black hole must be added to the mutables list since
+the TSOs on the list may contain pointers into the new generation.
+There is no need to clutter up the mutables list with black holes with
+empty blocking queues.
-\ToDo{MVars}
+\note{In a single-threaded system, entering a black hole indicates an
+infinite loop. In a concurrent system, entering a black hole
+indicates an infinite loop only if the hole is being entered by the
+same thread that originally entered the closure.}
\subsubsection{FetchMes}\label{sect:FETCHME}
\subsection{Unpointed Objects}
-A variable of unpointed type is always bound to a {\em value}, never to a {\em thunk}.
-For this reason, unpointed objects cannot be entered.
-
-A {\em value} may be:
-\begin{itemize}
-\item {\em Boxed}, i.e.~represented indirectly by a pointer to a heap object (e.g.~foreign objects, arrays); or
-\item {\em Unboxed}, i.e.~represented directly by a bit-pattern in one or more registers (e.g.~@Int#@ and @Float#@).
-\end{itemize}
-All {\em pointed} values are {\em boxed}.
+A variable of unpointed type is always bound to a \emph{value}, never
+to a \emph{thunk}. For this reason, unpointed objects cannot be
+entered.
\subsubsection{Immutable Objects}
\label{sect:ARR_WORDS1}
\begin{center}
\begin{tabular}{|c|c|c|c|}
\hline
-{\em Fixed Hdr} & {\em No of non-pointers} & {\em Non-pointers\ldots} \\ \hline
+\emph{Fixed Hdr} & \emph{No of non-pointers} & \emph{Non-pointers\ldots} \\ \hline
\end{tabular}
\end{center}
\begin{center}
\begin{tabular}{|c|c|c|c|}
\hline
-{\em Fixed Hdr} & {\em Mutable link} & {\em No of pointers} & {\em Pointers\ldots} \\ \hline
+\emph{Fixed Hdr} & \emph{Mutable link} & \emph{No of pointers} & \emph{Pointers\ldots} \\ \hline
\end{tabular}
\end{center}
The mutable link is present so that we can easily freeze and thaw an
\label{sect:MUTARR_PTRS}
\label{sect:MUTARR_PTRS_FROZEN}
-Some of these objects are {\em mutable}; they represent objects which
+Some of these objects are \emph{mutable}; they represent objects which
are explicitly mutated by Haskell code through the @ST@ monad.
They're not used for thunks which are updated precisely once.
Depending on the garbage collector, mutable closures may contain extra
\begin{description}
-\item[@ARR_WORDS@] is also used to represent {\em mutable} arrays of
+\item[@ARR_WORDS@] is also used to represent \emph{mutable} arrays of
bytes, words, floats, doubles, etc. It's possible to use the same
object type because even generational collectors don't need to
distinguish them.
\begin{center}
\begin{tabular}{|c|c|c|}
\hline
-{\em Fixed Hdr} & {\em Mutable link} & {\em Pointer} \\ \hline
+\emph{Fixed Hdr} & \emph{Mutable link} & \emph{Pointer} \\ \hline
\end{tabular}
\end{center}
\item[@MUTARR_PTRS@] is a mutable array of pointers.
-Such an array may be {\em frozen}, becoming an @SM_MUTARR_PTRS_FROZEN@, with a
+Such an array may be \emph{frozen}, becoming an @ARR_PTRS@, with a
different info-table.
\begin{center}
\begin{tabular}{|c|c|c|c|}
\hline
-{\em Fixed Hdr} & {\em Mutable link} & {\em No of ptrs} & {\em Pointers\ldots} \\ \hline
+\emph{Fixed Hdr} & \emph{Mutable link} & \emph{No of ptrs} & \emph{Pointers\ldots} \\ \hline
\end{tabular}
\end{center}
-\item[@MUTARR_PTRS_FROZEN@] is a frozen @MUTARR_PTRS@ closure.
-The garbage collector converts @MUTARR_PTRS_FROZEN@ to @ARR_PTRS@ as it removes them from
-the mutables list.
-
\end{description}
\begin{center}
\begin{tabular}{|l|l|l|l|}
\hline
-{\em Fixed header} & {\em Data} & {\em Free Routine} & {\em Foreign object link} \\
+\emph{Fixed header} & \emph{Data} & \emph{Free Routine} & \emph{Foreign object link} \\
\hline
\end{tabular}
\end{center}
evacuated), the free routine is called and the object is deleted from
the list.
+\subsubsection{MVars and IVars}
+\label{sect:MVAR}
+\label{sect:IVAR}
+
+\ToDo{MVars and IVars}
+
+
The remaining objects types are all administrative --- none of them may be entered.
A @BlockedFetch@ closure has the form:
\begin{center}
\begin{tabular}{|l|l|l|l|l|l|}\hline
-{\em Fixed header} & link & node & gtid & slot & weight \\ \hline
+\emph{Fixed header} & link & node & gtid & slot & weight \\ \hline
\end{tabular}
\end{center}
the closures in the pool. It may also choose to delete sparks from the pool.
\begin{center}
\begin{tabular}{|l|l|l|l|l|l|}\hline
-{\em Fixed header} & {\em Spark pool link} & {\em Sparked closure} \\ \hline
+\emph{Fixed header} & \emph{Spark pool link} & \emph{Sparked closure} \\ \hline
\end{tabular}
\end{center}
\begin{center}
\begin{tabular}{|l|l|l|l|l|l|}\hline
-{\em Info Pointer} & {\em Size} & {\em Slop Words} \\ \hline
+\emph{Info Pointer} & \emph{Size} & \emph{Slop Words} \\ \hline
\end{tabular}
\end{center}
``unpacked'' into machine registers and various other memory locations
to provide faster access.
-Single-threaded systems don't really {\em need\/} TSOs --- but they do
+Single-threaded systems don't really \emph{need\/} TSOs --- but they do
need some way to tell the storage manager about live roots so it is
convenient to use a single TSO to store the mutator state even in
single-threaded systems.
-Rather than manage TSOs' alloc/dealloc, etc., in some {\em ad hoc}
+Rather than manage TSOs' alloc/dealloc, etc., in some \emph{ad hoc}
way, we instead alloc/dealloc/etc them in the heap; then we can use
all the standard garbage-collection/fetching/flushing/etc machinery on
them. So that's why TSOs are ``heap objects,'' albeit very special
ones.
\begin{center}
\begin{tabular}{|l|l|}
- \hline {\em Fixed header}
+ \hline \emph{Fixed header}
\\ \hline @TSO_LINK@
-\\ \hline @TSO_WHATNEXT@
-\\ \hline @TSO_WHATNEXT_INFO@
-\\ \hline @TSO_STACK@
-\\ \hline {\em Exception Handlers}
-\\ \hline {\em Ticky Info}
-\\ \hline {\em Profiling Info}
-\\ \hline {\em Parallel Info}
-\\ \hline {\em GranSim Info}
-\\ \hline
+\\ \hline @TSO_STATE@
+\\ \hline \emph{Exception Handlers}
+\\ \hline \emph{Ticky Info}
+\\ \hline \emph{Profiling Info}
+\\ \hline \emph{Parallel Info}
+\\ \hline \emph{GranSim Info}
+\\ \hline
+\\
+ \emph{Stack}
+\\
+\\ \hline
\end{tabular}
\end{center}
The contents of a TSO are:
state (e.g.~all runnable, all sleeping, all blocked on the same black
hole, all blocked on the same MVar, etc.)
-\item A word (@TSO_WHATNEXT@) which is in suspended threads to record
- how to awaken it. This typically requires a program counter which is stored
- in the pointer @TSO_WHATNEXT_INFO@
-
-\item A pointer (@TSO_STACK@) to the top stack block.
+\item A word (@TSO_STATE@) which records the current state of a thread: running, runnable, blocked, etc.
\item Optional information for ``Ticky Ticky'' statistics: @TSO_STK_HWM@ is
the maximum number of words allocated to this thread.
@TSO_CCC@ is the current cost centre.
\item Optional information for parallel execution:
-\begin{itemize}
-
-\item The types of threads (@TSO_TYPE@):
-\begin{description}
-\item[@T_MAIN@] Must be executed locally.
-\item[@T_REQUIRED@] A required thread -- may be exported.
-\item[@T_ADVISORY@] An advisory thread -- may be exported.
-\item[@T_FAIL@] A failure thread -- may be exported.
-\end{description}
-
-\item I've no idea what else
-\end{itemize}
-
-\item Optional information for GranSim execution:
-\begin{itemize}
-\item locked
-\item sparkname
-\item started at
-\item exported
-\item basic blocks
-\item allocs
-\item exectime
-\item fetchtime
-\item fetchcount
-\item blocktime
-\item blockcount
-\item global sparks
-\item local sparks
-\item queue
-\item priority
-\item clock (gransim light only)
-\end{itemize}
-
-
-Here are the various queues for GrAnSim-type events.
-@
-Q_RUNNING
-Q_RUNNABLE
-Q_BLOCKED
-Q_FETCHING
-Q_MIGRATING
-@
+% \begin{itemize}
+%
+% \item The types of threads (@TSO_TYPE@):
+% \begin{description}
+% \item[@T_MAIN@] Must be executed locally.
+% \item[@T_REQUIRED@] A required thread -- may be exported.
+% \item[@T_ADVISORY@] An advisory thread -- may be exported.
+% \item[@T_FAIL@] A failure thread -- may be exported.
+% \end{description}
+%
+% \item I've no idea what else
+%
+% \end{itemize}
+%
+% \item Optional information for GranSim execution:
+% \begin{itemize}
+% \item locked
+% \item sparkname
+% \item started at
+% \item exported
+% \item basic blocks
+% \item allocs
+% \item exectime
+% \item fetchtime
+% \item fetchcount
+% \item blocktime
+% \item blockcount
+% \item global sparks
+% \item local sparks
+% \item queue
+% \item priority
+% \item clock (gransim light only)
+% \end{itemize}
+%
+%
+% Here are the various queues for GrAnSim-type events.
+%
+% Q_RUNNING
+% Q_RUNNABLE
+% Q_BLOCKED
+% Q_FETCHING
+% Q_MIGRATING
+%
\end{itemize}
\label{sect:STACK_OBJECT}
\label{sect:stacks}
+\ToDo{Merge this in with the section on TSOs}
+
These are ``stack objects,'' which are used in the threaded world as
the stack for each thread is allocated from the heap in smallish
chunks. (The stack in the sequential world is allocated outside of
\begin{center}
\begin{tabular}{|l|}
\hline
-{\em Fixed header}
+\emph{Fixed header}
\\ \hline
-{\em Link to next stack object (0 for last)}
+\emph{Link to next stack object (0 for last)}
\\ \hline
-{\em N, the payload size in words}
+\emph{N, the payload size in words}
\\ \hline
-{\em @Sp@ (byte offset from head of object)}
+\emph{@Sp@ (byte offset from head of object)}
\\ \hline
-{\em @Su@ (byte offset from head of object)}
+\emph{@Su@ (byte offset from head of object)}
\\ \hline
-{\em Payload (N words)}
+\emph{Payload (N words)}
\\ \hline
\end{tabular}
\end{center}
-\ToDo{Are stack objects on the mutable list?}
-
The stack grows downwards, towards decreasing
addresses. This makes it easier to print out the stack
when debugging, and it means that a return address is
of stack that the return address ``knows about'', namely the
activation record of the currently running function.
-\item Below each such activation record is a {\em pending-argument
+\item Below each such activation record is a \emph{pending-argument
section}, a chunk of
zero or more words that are the arguments to which the result
of the function should be applied. The return address does not
can consider update frames as a special case of a return address with
a well-defined activation record.
-\ToDo{Doesn't it {\em have} to be an update frame? After all, the arg
-satisfaction check is @Su - Sp >= ...@.}
-
\end{itemize}
The game plan is this. The garbage collector
\subsubsection{Activation records}\label{sect:activation-records}
-An {\em activation record} is a contiguous chunk of stack,
+
+An \emph{activation record} is a contiguous chunk of stack,
with a return address as its first word, followed by as many
data words as the return address ``knows about''. The return
address is actually a fully-fledged info pointer. It points
\begin{center}
\begin{tabular}{|l|l|l|l|l|l|l|l|l|l|l|}
\hline
-{\em Fixed header} & {\em No of pointers} & {\em Free} & $SP_0$ & \ldots & $SP_{n-1}$
+\emph{Fixed header} & \emph{No of pointers} & \emph{Free} & $SP_0$ & \ldots & $SP_{n-1}$
\\\hline
\end{tabular}
\end{center}
\item
-We use just one info table for {\em all\/} direct returns.
+We use just one info table for \emph{all\/} direct returns.
This introduces two problems:
\begin{enumerate}
\item How does the interpreter know what code to execute?
BCOs are removed from the symbol table, and the code will be garbage
collected some time later.
-A BCO represents a basic block of code - all entry points are at the
-beginning of a BCO, and it is impossible to jump into the middle of
-one. A BCO represents not only the code for a function, but also its
-closure; a BCO can be entered just like any other closure. Hugs
-performs lambda-lifting during compilation to byte-code, and each
+A BCO represents a basic block of code --- the (only) entry points is
+at the beginning of a BCO, and it is impossible to jump into the
+middle of one. A BCO represents not only the code for a function, but
+also its closure; a BCO can be entered just like any other closure.
+Hugs performs lambda-lifting during compilation to byte-code, and each
top-level combinator becomes a BCO in the heap.
-\ToDo{The phrase "all entry points..." suggests that BCOs have multiple
-entry points. If so, we need to say a lot more about it...}
\subsubsection{Thunks and partial applications}
> pushVar f
> SLIDE (m+1) |env|
> ENTER
-> cg (let{x1=rhs1; ... xm=rhsm in e) = do
+> cg (let {x1=rhs1; ... xm=rhsm} in e) = do
> ALLOC x1 |rhs1|, ... ALLOC xm |rhsm|
> build x1 rhs1, ... build xm rhsm
> cg e
> build x (C{a1, ... am}) = do
> pushUntaggedAtom am; ... pushUntaggedAtom a1
> PACK x C
+> -- A useful optimisation
+> build x ({v1, ... vm} \ {}. f{a1, ... am}) = do
+> pushVar am; ... pushVar a1
+> pushVar f
+> MKAP x m
> build x ({v1, ... vm} \ {}. e) = do
> pushVar vm; ... pushVar v1
> PUSHBCO (cgRhs ({v1, ... vm} \ {}. e))
\Arg{n+1} \ldots \Arg{m} are on top of the stack.
Note that the mapping of arguments \Arg{1} \ldots \Arg{n} to machine
-registers depends on the {\em kinds} of the arguments. For example,
+registers depends on the \emph{kinds} of the arguments. For example,
if the first argument is a Float, we might pass it in a different
register from if it is an Int. In fact, we might find that a given
architecture lets us pass varying numbers of arguments according to
cases to consider:
\begin{enumerate}
-\item {\em Known combinator (function with no free variables) and enough arguments.}
+\item \emph{Known combinator (function with no free variables) and enough arguments.}
A fast call can be made: push excess arguments onto stack and jump to
-function's {\em fast entry point} passing arguments in \Arg{1} \ldots
+function's \emph{fast entry point} passing arguments in \Arg{1} \ldots
\Arg{m}.
-The {\em fast entry point} is only called with exactly the right
+The \emph{fast entry point} is only called with exactly the right
number of arguments (in \Arg{1} \ldots \Arg{m}) so it can instantly
start doing useful work without first testing whether it has enough
registers or having to pop them off the stack first.
-\item {\em Known combinator and insufficient arguments.}
+\item \emph{Known combinator and insufficient arguments.}
A slow call can be made: push all arguments onto stack and jump to
-function's {\em slow entry point}.
+function's \emph{slow entry point}.
Any unpointed arguments which are pushed on the stack must be tagged.
This means pushing an extra word on the stack below the unpointed
%Todo: forward ref about tagging?
%Todo: picture?
-The {\em slow entry point} might be called with insufficient arguments
+The \emph{slow entry point} might be called with insufficient arguments
and so it must test whether there are enough arguments on the stack.
-This {\em argument satisfaction check} consists of checking that
+This \emph{argument satisfaction check} consists of checking that
@Su-Sp@ is big enough to hold all the arguments (including any tags).
\begin{itemize}
\end{itemize}
-\item {\em Known function closure (function with free variables) and enough arguments.}
+\item \emph{Known function closure (function with free variables) and enough arguments.}
A fast call can be made: push excess arguments onto stack and jump to
-function's {\em fast entry point} passing a pointer to closure in
+function's \emph{fast entry point} passing a pointer to closure in
\Arg{1} and arguments in \Arg{2} \ldots \Arg{m+1}.
Like the fast entry point for a combinator, the fast entry point for a
closure is used to access the free variables of the closure.
-\item {\em Known function closure and insufficient arguments.}
+\item \emph{Known function closure and insufficient arguments.}
A slow call can be made: push all arguments onto stack and jump to the
closure's slow entry point passing a pointer to the closure in \Arg{1}.
\Arg{2} \ldots \Arg{m+1} and jumps to the fast entry point.
-\item {\em Unknown function closure, thunk or constructor.}
+\item \emph{Unknown function closure, thunk or constructor.}
Sometimes, the function being called is not statically identifiable.
Consider, for example, the @compose@ function:
\Arg{1}. If \Arg{1} is passed on the stack, it is pushed on top of
the thunk for @(g x)@.
-The {\em entry code} for an updateable thunk (which must have arity 0)
+The \emph{entry code} for an updateable thunk (which must have arity 0)
pushes an update frame on the stack and starts executing the body of
the closure --- using \Arg{1} to access any free variables. This is
described in more detail in section~\ref{sect:data-updates}.
-The {\em entry code} for a non-updateable closure is just the
+The \emph{entry code} for a non-updateable closure is just the
closure's slow entry point.
\end{enumerate}
\subsection{Case expressions and return conventions}
\label{sect:return-conventions}
-The {\em evaluation} of a thunk is always initiated by
+The \emph{evaluation} of a thunk is always initiated by
a @case@ expression. For example:
@
case x of (a,b) -> E
Just a -> ...
@
Rather than pushing a return address before evaluating the scrutinee,
-@E@, the @case@ expression pushes (a pointer to) a {\em return
+@E@, the @case@ expression pushes (a pointer to) a \emph{return
vector}, a static table consisting of two code pointers: one for the
@Just@ alternative, and one for the @Nothing@ alternative.
\begin{itemize}
\item copies the free variables out of the thunk into registers or
onto the stack.
-\item pushes an {\em update frame} onto the stack.
+\item pushes an \emph{update frame} onto the stack.
An update frame is a small activation record consisting of
\begin{center}
\begin{tabular}{|l|l|l|}
\hline
-{\em Fixed header} & {\em Update Frame link} & {\em Updatee} \\
+\emph{Fixed header} & \emph{Update Frame link} & \emph{Updatee} \\
\hline
\end{tabular}
\end{center}
The update code:
\begin{itemize}
-\item overwrites the {\em updatee} with an indirection to \Arg{1};
+\item overwrites the \emph{updatee} with an indirection to \Arg{1};
\item loads @Su@ from the Update Frame link;
\item removes the update frame from the stack; and
\item enters \Arg{1}.
The storage manager detects that it needs to garbage collect the old
generation when the evaluator requests a garbage collection without
having moved the heap pointer since the last garbage collection. It
-is therefore important that the GC routines {\em not} move the heap
+is therefore important that the GC routines \emph{not} move the heap
pointer unless the heap check fails. This is different from what
happens in the current STG implementation.
\fi
-\part{Implementation}
-\section{Overview}
+\part{History}
-This part describes the inner workings of the major components of the system.
-Their external interfaces are described in the previous part.
+We're nuking the following:
-The major components of the system are:
\begin{itemize}
-\item The scheduler
-\item The loader
-\item The storage manager
-\item The machine code evaluator (compiled code)
-\item The bytecode evaluator (interpreted code)
-\end{itemize}
+\item
+ Two stacks
-\iffalse
+\item
+ Return in registers.
+ This lets us remove update code pointers from info tables,
+ removes the need for phantom info tables, simplifies
+ semi-tagging, etc.
-\section{Heap objects}
-\label{sect:heap-objects}
-\label{sect:fixed-header}
+\item
+ Threaded GC.
+ Careful analysis suggests that it doesn't buy us very much
+ and it is hard to work with.
-\ToDo{Fix this picture}
+ Eliminating threaded GCs eliminates the desire to share SMReps
+ so they are (once more) part of the Info table.
-\begin{figure}
-\begin{center}
-\input{closure}
-\end{center}
-\caption{A closure}
-\label{fig:closure}
-\end{figure}
+\item
+ RetReg.
+ Doesn't buy us anything on a register-poor architecture and
+ isn't so important if we have semi-tagging.
-Every {\em heap object} is a contiguous block
-of memory, consisting of a fixed-format {\em header} followed
-by zero or more {\em data words}.
+@
+ - Probably bad on register poor architecture
+ - Can avoid need to write return address to stack on reg rich arch.
+ - when a function does a small amount of work, doesn't
+ enter any other thunks and then returns.
+ eg entering a known constructor (but semitagging will catch this)
+ - Adds complications
+@
-\ToDo{I absolutely do not believe that every heap object has a header
-like this - ADR. I believe that they all have an info pointer but I
-see no readon why stack objects and unpointed heap objects would have
-an entry count since this will always be zero.}
+\item
+ Update in place
-The header consists of the following fields:
-\begin{itemize}
-\item A one-word {\em info pointer}, which points to
-the object's static {\em info table}.
-\item Zero or more {\em admin words} that support
-\begin{itemize}
-\item Profiling (notably a {\em cost centre} word).
- \note{We could possibly omit the cost centre word from some
- administrative objects.}
-\item Parallelism (e.g. GranSim keeps the object's global address here,
-though GUM keeps a separate hash table).
-\item Statistics (e.g. a word to track how many times a thunk is entered.).
+ This lets us drop CONST closures and CHARLIKE closures (assuming we
+ don't support Unicode). The only point of these closures was to
+ avoid updating with an indirection.
-We add a Ticky word to the fixed-header part of closures. This is
-used to indicate if a closure has been updated but not yet entered. It
-is set when the closure is updated and cleared when subsequently
-entered.
+ We also drop @MIN_UPD_SIZE@ --- all we need is space to insert an
+ indirection or a black hole.
-NB: It is {\em not} an ``entry count'', it is an
-``entries-after-update count.'' The commoning up of @CONST@,
-@CHARLIKE@ and @INTLIKE@ closures is turned off(?) if this is
-required. This has only been done for 2s collection.
+\item
+ STATIC SMReps are now called CONST
-\end{itemize}
-\end{itemize}
-
-Most of the RTS is completely insensitive to the number of admin words.
-The total size of the fixed header is @FIXED_HS@.
-
-Many heap objects contain fields allowing them to be inserted onto lists
-during evaluation or during garbage collection. The lists required by
-the evaluator and storage manager are as follows.
-
-\begin{itemize}
-\item 2 lists of threads: runnable threads and sleeping threads.
-
-\item The {\em static object list} is a list of all statically
-allocated objects which might contain pointers into the heap.
-(Section~\ref{sect:static-objects}.)
-
-\item The {\em updated thunk list} is a list of all thunks in the old
-generation which have been updated with an indirection.
-(Section~\ref{sect:IND_OLDGEN}.)
-
-\item The {\em mutables list} is a list of all other objects in the
-old generation which might contain pointers into the new generation.
-Most of the object on this list are ``mutable.''
-(Section~\ref{sect:mutables}.)
-
-\item The {\em Foreign Object list} is a list of all foreign objects
- which have not yet been deallocated. (Section~\ref{sect:FOREIGN}.)
-
-\item The {\em Spark pool} is a doubly(?) linked list of Spark objects
-maintained by the parallel system. (Section~\ref{sect:SPARK}.)
-
-\item The {\em Blocked Fetch list} (or
-lists?). (Section~\ref{sect:BLOCKED_FETCH}.)
-
-\item For each thread, there is a list of all update frames on the
-stack. (Section~\ref{sect:data-updates}.)
-
-
-\end{itemize}
-
-\ToDo{The links for these fields are usually inserted immediately
-after the fixed header except ...}
-
-\subsection{Info Tables}
-
-An {\em info table} is a contiguous block of memory, {\em laid out
-backwards}. That is, the first field in the list that follows
-occupies the highest memory address, and the successive fields occupy
-successive decreasing memory addresses.
-
-\begin{center}
-\begin{tabular}{|c|}
- \hline Parallelism Info
-\\ \hline Profile Info
-\\ \hline Debug Info
-\\ \hline Tag / Static reference table
-\\ \hline Storage manager layout info
-\\ \hline Closure type
-\\ \hline entry code
-\\ \vdots
-\end{tabular}
-\end{center}
-An info table has the following contents (working backwards in memory
-addresses):
-\begin{itemize}
-\item The {\em entry code} for the closure.
-This code appears literally as the (large) last entry in the
-info table, immediately preceded by the rest of the info table.
-An {\em info pointer} always points to the first byte of the entry code.
-
-\item A one-word {\em closure type field}, @INFO_TYPE@, identifies what kind
-of closure the object is. The various types of closure are described
-in Section~\ref{sect:closures}.
-In some configurations, some useful properties of
-closures (is it a HNF? can it be sparked?)
-are represented as high-order bits so they can be tested quickly.
-
-\item A single pointer or word --- the {\em storage manager info field},
-@INFO_SM@, contains auxiliary information describing the closure's
-precise layout, for the benefit of the garbage collector and the code
-that stuffs graph into packets for transmission over the network.
-
-\item A one-word {\em Tag/Static Reference Table} field, @INFO_SRT@.
-For data constructors, this field contains the constructor tag, in the
-range $0..n-1$ where $n$ is the number of constructors. For all other
-objects it contains a pointer to a table which enables the garbage
-collector to identify all accessible code and CAFs. They are fully
-described in Section~\ref{sect:srt}.
-
-\item {\em Profiling info\/}
-
-Closure category records are attached to the info table of the
-closure. They are declared with the info table. We put pointers to
-these ClCat things in info tables. We need these ClCat things because
-they are mutable, whereas info tables are immutable. Hashing will map
-similar categories to the same hash value allowing statistics to be
-grouped by closure category.
-
-Cost Centres and Closure Categories are hashed to provide indexes
-against which arbitrary information can be stored. These indexes are
-memoised in the appropriate cost centre or category record and
-subsequent hashes avoided by the index routine (it simply returns the
-memoised index).
-
-There are different features which can be hashed allowing information
-to be stored for different groupings. Cost centres have the cost
-centre recorded (using the pointer), module and group. Closure
-categories have the closure description and the type
-description. Records with the same feature will be hashed to the same
-index value.
-
-The initialisation routines, @init_index_<feature>@, allocate a hash
-table in which the cost centre / category records are stored. The
-lower bound for the table size is taken from @max_<feature>_no@. They
-return the actual table size used (the next power of 2). Unused
-locations in the hash table are indicated by a 0 entry. Successive
-@init_index_<feature>@ calls just return the actual table size.
-
-Calls to @index_<feature>@ will insert the cost centre / category
-record in the @<feature>@ hash table, if not already inserted. The hash
-index is memoised in the record and returned.
-
-CURRENTLY ONLY ONE MEMOISATION SLOT IS AVILABLE IN EACH RECORD SO
-HASHING CAN ONLY BE DONE ON ONE FEATURE FOR EACH RECORD. This can be
-easily relaxed at the expense of extra memoisation space or continued
-rehashing.
-
-The initialisation routines must be called before initialisation of
-the stacks and heap as they require to allocate storage. It is also
-expected that the caller may want to allocate additional storage in
-which to store profiling information based on the return table size
-value(s).
-
-\begin{center}
-\begin{tabular}{|l|}
- \hline Hash Index
-\\ \hline Selected
-\\ \hline Kind
-\\ \hline Description String
-\\ \hline Type String
-\\ \hline
-\end{tabular}
-\end{center}
-
-\begin{description}
-\item[Hash Index] Memoised copy
-\item[Selected]
- Is this category selected (-1 == not memoised, selected? 0 or 1)
-\item[Kind]
-One of the following values (defined in CostCentre.lh):
-
-\begin{description}
-\item[@CON_K@]
-A constructor.
-\item[@FN_K@]
-A literal function.
-\item[@PAP_K@]
-A partial application.
-\item[@THK_K@]
-A thunk, or suspension.
-\item[@BH_K@]
-A black hole.
-\item[@ARR_K@]
-An array.
-\item[@ForeignObj_K@]
-A Foreign object (non-Haskell heap resident).
-\item[@SPT_K@]
-The Stable Pointer table. (There should only be one of these but it
-represents a form of weak space leak since it can't shrink to meet
-non-demand so it may be worth watching separately? ADR)
-\item[@INTERNAL_KIND@]
-Something internal to the runtime system.
-\end{description}
-
-
-\item[Description] Source derived string detailing closure description.
-\item[Type] Source derived string detailing closure type.
-\end{description}
-
-\item {\em Parallelism info\/}
-\ToDo{}
-
-\item {\em Debugging info\/}
-\ToDo{}
-
-\end{itemize}
-
-
-%-----------------------------------------------------------------------------
-\subsection{Kinds of Heap Object}
-\label{sect:closures}
-
-Heap objects can be classified in several ways, but one useful one is
-this:
-\begin{itemize}
-\item
-{\em Static closures} occupy fixed, statically-allocated memory
-locations, with globally known addresses.
-
-\item
-{\em Dynamic closures} are individually allocated in the heap.
-
-\item
-{\em Stack closures} are closures allocated within a thread's stack
-(which is itself a heap object). Unlike other closures, there are
-never any pointers to stack closures. Stack closures are discussed in
-Section~\ref{sect:stacks}.
-
-\end{itemize}
-A second useful classification is this:
-\begin{itemize}
-\item
-{\em Executive objects}, such as thunks and data constructors,
-participate directly in a program's execution. They can be subdivided into
-three kinds of objects according to their type:
-\begin{itemize}
-\item
-{\em Pointed objects}, represent values of a {\em pointed} type
-(<.pointed types launchbury.>) --i.e.~a type that includes $\bottom$ such as @Int@ or @Int# -> Int#@.
-
-\item {\em Unpointed objects}, represent values of a {\em unpointed} type --i.e.~a type that does not include $\bottom$ such as @Int#@ or @Array#@.
-
-\item {\em Activation frames}, represent ``continuations''. They are
-always stored on the stack and are never pointed to by heap objects or
-passed as arguments. \note{It's not clear if this will still be true
-once we support speculative evaluation.}
-
-\end{itemize}
-
-\item {\em Administrative objects}, such as stack objects and thread
-state objects, do not represent values in the original program.
-\end{itemize}
-
-Only pointed objects can be entered. All pointed objects share a
-common header format: the ``pointed header''; while all unpointed
-objects share a common header format: the ``unpointed header''.
-\ToDo{Describe the difference and update the diagrams to mention
-an appropriate header type.}
-
-This section enumerates all the kinds of heap objects in the system.
-Each is identified by a distinct @INFO_TYPE@ tag in its info table.
-
-\ToDo{Check this table very carefully}
-
-\begin{tabular}{|l|l|l|l|l|l|l|l|l|l|l|}
-\hline
-
-closure kind & HNF & UPD & NS & STA & THU & MUT & UPT & BH & IND & Section \\
-
-\hline
-{\em Pointed} \\
-\hline
-
-@CONSTR@ & 1 & & 1 & & & & & & & \ref{sect:CONSTR} \\
-@CONSTR_STATIC@ & 1 & & 1 & 1 & & & & & & \ref{sect:CONSTR} \\
-@CONSTR_STATIC_NOCAF@ & 1 & & 1 & 1 & & & & & & \ref{sect:CONSTR} \\
-
-@FUN@ & 1 & & ? & & & & & & & \ref{sect:FUN} \\
-@FUN_STATIC@ & 1 & & ? & 1 & & & & & & \ref{sect:FUN} \\
-
-@THUNK@ & & 1 & & & 1 & & & & & \ref{sect:THUNK} \\
-@THUNK_STATIC@ & & 1 & & 1 & 1 & & & & & \ref{sect:THUNK} \\
-@THUNK_SELECTOR@ & & 1 & 1 & & 1 & & & & & \ref{sect:THUNK_SEL} \\
-
-@BCO@ & 1 & & 1 & & & & & & & \ref{sect:BCO} \\
-@BCO_CAF@ & & 1 & & & 1 & & & & & \ref{sect:BCO} \\
-
-@AP@ & & 1 & & & 1 & & & & & \ref{sect:AP} \\
-@PAP@ & 1 & & 1 & & & & & & & \ref{sect:PAP} \\
-
-@IND@ & ? & & ? & & ? & & & & 1 & \ref{sect:IND} \\
-@IND_OLDGEN@ & ? & & ? & & ? & & & & 1 & \ref{sect:IND} \\
-@IND_PERM@ & ? & & ? & & ? & & & & 1 & \ref{sect:IND} \\
-@IND_OLDGEN_PERM@ & ? & & ? & & ? & & & & 1 & \ref{sect:IND} \\
-@IND_STATIC@ & ? & & ? & 1 & ? & & & & 1 & \ref{sect:IND} \\
-
-\hline
-{\em Unpointed} \\
-\hline
-
-
-@ARR_WORDS@ & 1 & & 1 & & & & 1 & & & \ref{sect:ARR_WORDS1},\ref{sect:ARR_WORDS2} \\
-@ARR_PTRS@ & 1 & & 1 & & & & 1 & & & \ref{sect:ARR_PTRS} \\
-@MUTVAR@ & 1 & & 1 & & & 1 & 1 & & & \ref{sect:MUTVAR} \\
-@MUTARR_PTRS@ & 1 & & 1 & & & 1 & 1 & & & \ref{sect:MUTARR_PTRS} \\
-@MUTARR_PTRS_FROZEN@ & 1 & & 1 & & & 1 & 1 & & & \ref{sect:MUTARR_PTRS_FROZEN} \\
-
-@FOREIGN@ & 1 & & 1 & & & & 1 & & & \ref{sect:FOREIGN} \\
-
-@BH@ & & 1 & 1 & & ? & ? & & 1 & ? & \ref{sect:BH} \\
-@MVAR@ & 1 & & 1 & & & & & & & \ref{sect:MVAR} \\
-@IVAR@ & 1 & & 1 & & & & & & & \ref{sect:IVAR} \\
-@FETCHME@ & 1 & & 1 & & & & & & & \ref{sect:FETCHME} \\
-\hline
-\end{tabular}
-
-Activation frames do not live (directly) on the heap --- but they have
-a similar organisation. The classification bits are all zero in
-activation frames.
-
-\begin{tabular}{|l|l|}\hline
-closure kind & Section \\ \hline
-@RET_SMALL@ & \ref{sect:activation-records} \\
-@RET_VEC_SMALL@ & \ref{sect:activation-records} \\
-@RET_BIG@ & \ref{sect:activation-records} \\
-@RET_VEC_BIG@ & \ref{sect:activation-records} \\
-@UPDATE_FRAME@ & \ref{sect:activation-records} \\
-\hline
-\end{tabular}
-
-There are also a number of administrative objects. The classification bits are
-all zero in administrative objects.
-
-\begin{tabular}{|l|l|}\hline
-closure kind & Section \\ \hline
-@TSO@ & \ref{sect:TSO} \\
-@STACK_OBJECT@ & \ref{sect:STACK_OBJECT} \\
-@STABLEPTR_TABLE@ & \ref{sect:STABLEPTR_TABLE} \\
-@SPARK_OBJECT@ & \ref{sect:SPARK} \\
-@BLOCKED_FETCH@ & \ref{sect:BLOCKED_FETCH} \\
-\hline
-\end{tabular}
-
-\ToDo{I guess the parallel system has something like a stable ptr
-table. Is there any opportunity for sharing code/data structures
-here?}
-
-
-\subsection{Classification bits}
-
-The top bits of the @INFO_TYPE@ tag tells what sort of animal the
-closure is.
-
-\begin{tabular}{|l|l|l|} \hline
-Abbrev & Bit & Interpretation \\ \hline
-HNF & 0 & 1 $\Rightarrow$ Head normal form \\
-UPD & 4 & 1 $\Rightarrow$ May be updated (inconsistent with being a HNF) \\
-NS & 1 & 1 $\Rightarrow$ Don't spark me (Any HNF will have this set to 1)\\
-STA & 2 & 1 $\Rightarrow$ This is a static closure \\
-THU & 8 & 1 $\Rightarrow$ Is a thunk \\
-MUT & 3 & 1 $\Rightarrow$ Has mutable pointer fields \\
-UPT & 5 & 1 $\Rightarrow$ Has an unpointed type (eg a primitive array) \\
-BH & 6 & 1 $\Rightarrow$ Is a black hole \\
-IND & 7 & 1 $\Rightarrow$ Is an indirection \\
-\hline
-\end{tabular}
-
-Updatable structures (@_UP@) are thunks that may be shared. Primitive
-arrays (@_BM@ -- Big Mothers) are structures that are always held
-in-memory (basically extensions of a closure). Because there may be
-offsets into these arrays, a primitive array cannot be handled as a
-FetchMe in the parallel system, but must be shipped in its entirety if
-its parent closure is shipped.
-
-The other bits in the info-type field simply give a unique bit-pattern
-to identify the closure type.
-
-\iffalse
-@
-#define _NF 0x0001 /* Normal form */
-#define _NS 0x0002 /* Don't spark */
-#define _ST 0x0004 /* Is static */
-#define _MU 0x0008 /* Is mutable */
-#define _UP 0x0010 /* Is updatable (but not mutable) */
-#define _BM 0x0020 /* Is a "primitive" array */
-#define _BH 0x0040 /* Is a black hole */
-#define _IN 0x0080 /* Is an indirection */
-#define _TH 0x0100 /* Is a thunk */
-
-
-
-SPEC
-SPEC_N SPEC | _NF | _NS
-SPEC_S SPEC | _TH
-SPEC_U SPEC | _UP | _TH
-
-GEN
-GEN_N GEN | _NF | _NS
-GEN_S GEN | _TH
-GEN_U GEN | _UP | _TH
-
-DYN _NF | _NS
-TUPLE _NF | _NS | _BM
-DATA _NF | _NS | _BM
-MUTUPLE _NF | _NS | _MU | _BM
-IMMUTUPLE _NF | _NS | _BM
-STATIC _NS | _ST
-CONST _NF | _NS
-CHARLIKE _NF | _NS
-INTLIKE _NF | _NS
-
-BH _NS | _BH
-BH_N BH
-BH_U BH | _UP
-
-BQ _NS | _MU | _BH
-IND _NS | _IN
-CAF _NF | _NS | _ST | _IN
-
-FM
-FETCHME FM | _MU
-FMBQ FM | _MU | _BH
-
-TSO _MU
-
-STKO
-STKO_DYNAMIC STKO | _MU
-STKO_STATIC STKO | _ST
-
-SPEC_RBH _NS | _MU | _BH
-GEN_RBH _NS | _MU | _BH
-BF _NS | _MU | _BH
-INTERNAL
-
-@
-\fi
-
-Notes:
-
-An indirection either points to HNF (post update); or is result of
-overwriting a FetchMe, in which case the thing fetched is either
-under evaluation (BH), or by now an HNF. Thus, indirections get NoSpark flag.
-
-
-
-\subsection{Hugs Objects}
-
-\subsubsection{Byte-Code Objects}
-\label{sect:BCO}
-
-A Byte-Code Object (BCO) is a container for a a chunk of byte-code,
-which can be executed by Hugs. The byte-code represents a
-supercombinator in the program: when hugs compiles a module, it
-performs lambda lifting and each resulting supercombinator becomes a
-byte-code object in the heap.
-
-There are two kinds of BCO: a standard @BCO@ which has an arity of one
-or more, and a @BCO_CAF@ which takes no arguments and can be updated.
-When a @BCO_CAF@ is updated, the code is thrown away!
-
-The semantics of BCOs are described in Section
-\ref{sect:hugs-heap-objects}. A BCO has the following structure:
-
-\begin{center}
-\begin{tabular}{|l|l|l|l|l|l|}
-\hline
-\emph{Fixed Header} & \emph{Layout} & \emph{Offset} & \emph{Size} &
-\emph{Literals} & \emph{Byte code} \\
-\hline
-\end{tabular}
-\end{center}
-
-\noindent where:
-\begin{itemize}
-\item The entry code is a static code fragment/info table that
-returns to the scheduler to invoke Hugs (Section
-\ref{sect:ghc-to-hugs-switch}).
-\item \emph{Layout} contains the number of pointer literals in the
-\emph{Literals} field.
-\item \emph{Offset} is the offset to the byte code from the start of
-the object.
-\item \emph{Size} is the number of words of byte code in the object.
-\item \emph{Literals} contains any pointer and non-pointer literals used in
-the byte-codes (including jump addresses), pointers first.
-\item \emph{Byte code} contains \emph{Size} words of non-pointer byte
-code.
-\end{itemize}
-
-\subsection{Pointed Objects}
-
-All pointed objects can be entered.
-
-\subsubsection{Function closures}\label{sect:FUN}
-
-Function closures represent lambda abstractions. For example,
-consider the top-level declaration:
-@
- f = \x -> let g = \y -> x+y
- in g x
-@
-Both @f@ and @g@ are represented by function closures. The closure
-for @f@ is {\em static} while that for @g@ is {\em dynamic}.
-
-The layout of a function closure is as follows:
-\begin{center}
-\begin{tabular}{|l|l|l|l|}\hline
-{\em Fixed header} & {\em Pointers} & {\em Non-pointers} \\ \hline
-\end{tabular}
-\end{center}
-The data words (pointers and non-pointers) are the free variables of
-the function closure.
-The number of pointers
-and number of non-pointers are stored in the @INFO_SM@ word, in the least significant
-and most significant half-word respectively.
-
-There are several different sorts of function closure, distinguished
-by their @INFO_TYPE@ field:
-\begin{itemize}
-\item @FUN@: a vanilla, dynamically allocated on the heap.
-
-\item $@FUN_@p@_@np$: to speed up garbage collection a number of
-specialised forms of @FUN@ are provided, for particular $(p,np)$ pairs,
-where $p$ is the number of pointers and $np$ the number of non-pointers.
-
-\item @FUN_STATIC@. Top-level, static, function closures (such as
-@f@ above) have a different
-layout than dynamic ones:
-\begin{center}
-\begin{tabular}{|l|l|l|}\hline
-{\em Fixed header} & {\em Static object link} \\ \hline
-\end{tabular}
-\end{center}
-Static function closures have no free variables. (However they may refer to other
-static closures; these references are recorded in the function closure's SRT.)
-They have one field that is not present in dynamic closures, the {\em static object
-link} field. This is used by the garbage collector in the same way that to-space
-is, to gather closures that have been determined to be live but that have not yet
-been scavenged.
-\note{Static function closures that have no static references, and hence
-a null SRT pointer, don't need the static object link field. Is it worth
-taking advantage of this? See @CONSTR_STATIC_NOCAF@.}
-\end{itemize}
-
-Each lambda abstraction, $f$, in the STG program has its own private info table.
-The following labels are relevant:
-\begin{itemize}
-\item $f$@_info@ is $f$'s info table.
-\item $f$@_entry@ is $f$'s slow entry point (i.e. the entry code of its
-info table; so it will label the same byte as $f$@_info@).
-\item $f@_fast_@k$ is $f$'s fast entry point. $k$ is the number of arguments
-$f$ takes; encoding this number in the fast-entry label occasionally catches some nasty
-code-generation errors.
-\end{itemize}
-
-\subsubsection{Data Constructors}\label{sect:CONSTR}
-
-Data-constructor closures represent values constructed with
-algebraic data type constructors.
-The general layout of data constructors is the same as that for function
-closures. That is
-\begin{center}
-\begin{tabular}{|l|l|l|l|}\hline
-{\em Fixed header} & {\em Pointers} & {\em Non-pointers} \\ \hline
-\end{tabular}
-\end{center}
-
-The SRT pointer in a data constructor's info table is used for the
-constructor tag, since a constructor never has any static references.
-
-There are several different sorts of constructor:
-\begin{itemize}
-\item @CONSTR@: a vanilla, dynamically allocated constructor.
-\item @CONSTR_@$p$@_@$np$: just like $@FUN_@p@_@np$.
-\item @CONSTR_INTLIKE@.
-A dynamically-allocated heap object that looks just like an @Int@. The
-garbage collector checks to see if it can common it up with one of a fixed
-set of static int-like closures, thus getting it out of the dynamic heap
-altogether.
-
-\item @CONSTR_CHARLIKE@: same deal, but for @Char@.
-
-\item @CONSTR_STATIC@ is similar to @FUN_STATIC@, with the complication that
-the layout of the constructor must mimic that of a dynamic constructor,
-because a static constructor might be returned to some code that unpacks it.
-So its layout is like this:
-\begin{center}
-\begin{tabular}{|l|l|l|l|l|}\hline
-{\em Fixed header} & {\em Pointers} & {\em Non-pointers} & {\em Static object link}\\ \hline
-\end{tabular}
-\end{center}
-The static object link, at the end of the closure, serves the same purpose
-as that for @FUN_STATIC@. The pointers in the static constructor can point
-only to other static closures.
-
-The static object link occurs last in the closure so that static
-constructors can store their data fields in exactly the same place as
-dynamic constructors.
-
-\item @CONSTR_STATIC_NOCAF@. A statically allocated data constructor
-that guarantees not to point (directly or indirectly) to any CAF
-(section~\ref{sect:CAF}). This means it does not need a static object
-link field. Since we expect that there might be quite a lot of static
-constructors this optimisation makes sense. Furthermore, the @NOCAF@
-tag allows the compiler to indicate that no CAFs can be reached
-anywhere {\em even indirectly}.
-
-
-\end{itemize}
-
-For each data constructor $Con$, two
-info tables are generated:
-\begin{itemize}
-\item $Con$@_info@ labels $Con$'s dynamic info table,
-shared by all dynamic instances of the constructor.
-\item $Con$@_static@ labels $Con$'s static info table,
-shared by all static instances of the constructor.
-\end{itemize}
-
-\subsubsection{Thunks}
-\label{sect:THUNK}
-\label{sect:THUNK_SEL}
-
-A thunk represents an expression that is not obviously in head normal
-form. For example, consider the following top-level definitions:
-@
- range = between 1 10
- f = \x -> let ys = take x range
- in sum ys
-@
-Here the right-hand sides of @range@ and @ys@ are both thunks; the former
-is static while the latter is dynamic.
-
-The layout of a thunk is the same as that for a function closure,
-except that it may have some words of ``slop'' at the end to make sure
-that it has
-at least @MIN_UPD_PAYLOAD@ words in addition to its
-fixed header.
-\begin{center}
-\begin{tabular}{|l|l|l|l|l|}\hline
-{\em Fixed header} & {\em Pointers} & {\em Non-pointers} & {\em Slop} \\ \hline
-\end{tabular}
-\end{center}
-The @INFO_SM@ word contains the same information as for function
-closures; that is, number of pointers and number of non-pointers (excluding slop).
-
-A thunk differs from a function closure in that it can be updated.
-
-There are several forms of thunk:
-\begin{itemize}
-\item @THUNK@: a vanilla, dynamically allocated thunk.
-The garbage collection code for thunks whose
-pointer + non-pointer words is less than @MIN_UPD_PAYLOAD@ differs from
-that for function closures and data constructors, because the GC code
-has to account for the slop.
-\item $@THUNK_@p@_@np$. Similar comments apply.
-\item @THUNK_STATIC@. A static thunk is also known as
-a {\em constant applicative form}, or {\em CAF}.
-
-\begin{center}
-\begin{tabular}{|l|l|l|l|l|}\hline
-{\em Fixed header} & {\em Pointers} & {\em Non-pointers} & {\em Slop} & {\em Static object link}\\ \hline
-\end{tabular}
-\end{center}
-
-\item @THUNK_SELECTOR@ is a (dynamically allocated) thunk
-whose entry code performs a simple selection operation from
-a data constructor drawn from a single-constructor type. For example,
-the thunk
-@
- x = case y of (a,b) -> a
-@
-is a selector thunk. A selector thunk is laid out like this:
-\begin{center}
-\begin{tabular}{|l|l|l|l|}\hline
-{\em Fixed header} & {\em Selectee pointer} \\ \hline
-\end{tabular}
-\end{center}
-The @INFO_SM@ word contains the byte offset of the desired word in
-the selectee. Note that this is different from all other thunks.
-
-The garbage collector ``peeks'' at the selectee's
-tag (in its info table). If it is evaluated, then it goes ahead and do
-the selection, and then behaves just as if the selector thunk was an
-indirection to the selected field.
-If it is not
-evaluated, it treats the selector thunk like any other thunk of that
-shape. [Implementation notes.
-Copying: only the evacuate routine needs to be special.
-Compacting: only the PRStart (marking) routine needs to be special.]
-\end{itemize}
-
-
-The only label associated with a thunk is its info table:
-\begin{description}
-\item[$f$@_info@] is $f$'s info table.
-\end{description}
-
-
-\subsubsection{Partial applications (PAPs)}\label{sect:PAP}
-
-A partial application (PAP) represents a function applied to too few arguments.
-It is only built as a result of updating after an argument-satisfaction
-check failure. A PAP has the following shape:
-\begin{center}
-\begin{tabular}{|l|l|l|l|}\hline
-{\em Fixed header} & {\em No of arg words} & {\em Function closure} & {\em Arg stack} \\ \hline
-\end{tabular}
-\end{center}
-The ``arg stack'' is a copy of the chunk of stack above the update
-frame; ``no of arg words'' tells how many words it consists of. The
-function closure is (a pointer to) the closure for the function whose
-argument-satisfaction check failed.
-
-There is just one standard form of PAP with @INFO_TYPE@ = @PAP@.
-There is just one info table too, called @PAP_info@.
-Its entry code simply copies the arg stack chunk back on top of the
-stack and enters the function closure. (It has to do a stack overflow test first.)
-
-PAPs are also used to implement Hugs functions (where the arguments are free variables).
-PAPs generated by Hugs can be static.
-
-\subsubsection{@AP@ objects}
-\label{sect:AP}
-
-@AP@ objects are used to represent thunks built by Hugs. The only distintion between
-an @AP@ and a @PAP@ is that an @AP@ is updateable.
-
-\begin{center}
-\begin{tabular}{|l|l|l|l|}
-\hline
-\emph{Fixed Header} & {\em No of arg words} & {\em Function closure} & {\em Arg stack} \\
-\hline
-\end{tabular}
-\end{center}
-
-The entry code pushes an update frame, copies the arg stack chunk on
-top of the stack, and enters the function closure. (It has to do a
-stack overflow test first.)
-
-The ``arg stack'' is a block of (tagged) arguments representing the
-free variables of the thunk; ``no of arg words'' tells how many words
-it consists of. The function closure is (a pointer to) the closure
-for the thunk. The argument stack may be empty if the thunk has no
-free variables.
-
-
-\subsubsection{Indirections}
-\label{sect:IND}
-\label{sect:IND_OLDGEN}
-
-Indirection closures just point to other closures. They are introduced
-when a thunk is updated to point to its value.
-The entry code for all indirections simply enters the closure it points to.
-
-There are several forms of indirection:
-\begin{description}
-\item[@IND@] is the vanilla, dynamically-allocated indirection.
-It is removed by the garbage collector. It has the following
-shape:
-\begin{center}
-\begin{tabular}{|l|l|l|}\hline
-{\em Fixed header} & {\em Target closure} \\ \hline
-\end{tabular}
-\end{center}
-
-\item[@IND_OLDGEN@] is the indirection used to update an old-generation
-thunk. Its shape is like this:
-\begin{center}
-\begin{tabular}{|l|l|l|}\hline
-{\em Fixed header} & {\em Mutable link field} & {\em Target closure} \\ \hline
-\end{tabular}
-\end{center}
-It contains a {\em mutable link field} that is used to string together
-all old-generation indirections that might have a pointer into
-the new generation.
-
-
-\item[@IND_PERMANENT@ and @IND_OLDGEN_PERMANENT@.]
-for lexical profiling, it is necessary to maintain cost centre
-information in an indirection, so ``permanent indirections'' are
-retained forever. Otherwise they are just like vanilla indirections.
-\note{If a permanent indirection points to another permanent
-indirection or a @CONST@ closure, it is possible to elide the indirection
-since it will have no effect on the profiler.}
-\note{Do we still need @IND@ and @IND_OLDGEN@
-in the profiling build, or can we just make
-do with one pair whose behaviour changes when profiling is built?}
-
-\item[@IND_STATIC@] is used for overwriting CAFs when they have been
-evaluated. Static indirections are not removed by the garbage
-collector; and are statically allocated outside the heap (and should
-stay there). Their static object link field is used just as for
-@FUN_STATIC@ closures.
-
-\begin{center}
-\begin{tabular}{|l|l|l|}
-\hline
-{\em Fixed header} & {\em Target closure} & {\em Static object link} \\
-\hline
-\end{tabular}
-\end{center}
-
-\end{description}
-
-\subsubsection{Activation Records}
-
-Activation records are parts of the stack described by return address
-info tables (closures with @INFO_TYPE@ values of @RET_SMALL@,
-@RET_VEC_SMALL@, @RET_BIG@ and @RET_VEC_BIG@). They are described in
-section~\ref{sect:activation-records}.
-
-
-\subsubsection{Black holes, MVars and IVars}
-\label{sect:BH}
-\label{sect:MVAR}
-\label{sect:IVAR}
-
-Black hole closures are used to overwrite closures currently being
-evaluated. They inform the garbage collector that there are no live
-roots in the closure, thus removing a potential space leak.
-
-Black holes also become synchronization points in the threaded world.
-They contain a pointer to a list of blocked threads to be awakened
-when the black hole is updated (or @NULL@ if the list is empty).
-\begin{center}
-\begin{tabular}{|l|l|l|}
-\hline
-{\em Fixed header} & {\em Mutable link} & {\em Blocked thread link} \\
-\hline
-\end{tabular}
-\end{center}
-The {\em Blocked thread link} points to the TSO of the first thread
-waiting for the value of this thunk. All subsequent TSOs in the list
-are linked together using their @TSO_LINK@ field.
-
-When the blocking queue is non-@NULL@, the black hole must be added to
-the mutables list since the TSOs on the list may contain pointers into
-the new generation. There is no need to clutter up the mutables list
-with black holes with empty blocking queues.
-
-\ToDo{MVars}
-
-
-\subsubsection{FetchMes}\label{sect:FETCHME}
-
-In the parallel systems, FetchMes are used to represent pointers into
-the global heap. When evaluated, the value they point to is read from
-the global heap.
-
-\ToDo{Describe layout}
-
-
-\subsection{Unpointed Objects}
-
-A variable of unpointed type is always bound to a {\em value}, never to a {\em thunk}.
-For this reason, unpointed objects cannot be entered.
-
-A {\em value} may be:
-\begin{itemize}
-\item {\em Boxed}, i.e.~represented indirectly by a pointer to a heap object (e.g.~foreign objects, arrays); or
-\item {\em Unboxed}, i.e.~represented directly by a bit-pattern in one or more registers (e.g.~@Int#@ and @Float#@).
-\end{itemize}
-All {\em pointed} values are {\em boxed}.
-
-\subsubsection{Immutable Objects}
-\label{sect:ARR_WORDS1}
-\label{sect:ARR_PTRS}
-
-\begin{description}
-\item[@ARR_WORDS@] is a variable-sized object consisting solely of
-non-pointers. It is used for arrays of all
-sorts of things (bytes, words, floats, doubles... it doesn't matter).
-\begin{center}
-\begin{tabular}{|c|c|c|c|}
-\hline
-{\em Fixed Hdr} & {\em No of non-pointers} & {\em Non-pointers\ldots} \\ \hline
-\end{tabular}
-\end{center}
-
-\item[@ARR_PTRS@] is an immutable, variable sized array of pointers.
-\begin{center}
-\begin{tabular}{|c|c|c|c|}
-\hline
-{\em Fixed Hdr} & {\em Mutable link} & {\em No of pointers} & {\em Pointers\ldots} \\ \hline
-\end{tabular}
-\end{center}
-The mutable link is present so that we can easily freeze and thaw an
-array (by changing the header and adding/removing the array to the
-mutables list).
-
-\end{description}
-
-\subsubsection{Mutable Objects}
-\label{sect:mutables}
-\label{sect:ARR_WORDS2}
-\label{sect:MUTVAR}
-\label{sect:MUTARR_PTRS}
-\label{sect:MUTARR_PTRS_FROZEN}
-
-Some of these objects are {\em mutable}; they represent objects which
-are explicitly mutated by Haskell code through the @ST@ monad.
-They're not used for thunks which are updated precisely once.
-Depending on the garbage collector, mutable closures may contain extra
-header information which allows a generational collector to implement
-the ``write barrier.''
-
-\begin{description}
-
-\item[@ARR_WORDS@] is also used to represent {\em mutable} arrays of
-bytes, words, floats, doubles, etc. It's possible to use the same
-object type because even generational collectors don't need to
-distinguish them.
-
-\item[@MUTVAR@] is a mutable variable.
-\begin{center}
-\begin{tabular}{|c|c|c|}
-\hline
-{\em Fixed Hdr} & {\em Mutable link} & {\em Pointer} \\ \hline
-\end{tabular}
-\end{center}
-
-\item[@MUTARR_PTRS@] is a mutable array of pointers.
-Such an array may be {\em frozen}, becoming an @SM_MUTARR_PTRS_FROZEN@, with a
-different info-table.
-\begin{center}
-\begin{tabular}{|c|c|c|c|}
-\hline
-{\em Fixed Hdr} & {\em Mutable link} & {\em No of ptrs} & {\em Pointers\ldots} \\ \hline
-\end{tabular}
-\end{center}
-
-\item[@MUTARR_PTRS_FROZEN@] is a frozen @MUTARR_PTRS@ closure.
-The garbage collector converts @MUTARR_PTRS_FROZEN@ to @ARR_PTRS@ as it removes them from
-the mutables list.
-
-\end{description}
-
-
-\subsubsection{Foreign Objects}\label{sect:FOREIGN}
-
-Here's what a ForeignObj looks like:
-
-\begin{center}
-\begin{tabular}{|l|l|l|l|}
-\hline
-{\em Fixed header} & {\em Data} & {\em Free Routine} & {\em Foreign object link} \\
-\hline
-\end{tabular}
-\end{center}
-
-The @FreeRoutine@ is a reference to the finalisation routine to call
-when the @ForeignObj@ becomes garbage. If @freeForeignObject@ is
-called on a Foreign Object, the @FreeRoutine@ is set to zero and the
-garbage collector will not attempt to call @FreeRoutine@ when the
-object becomes garbage.
-
-The Foreign object link is a link to the next foreign object in the
-list. This list is traversed at the end of garbage collection: if an
-object is about to be deallocated (e.g.~it was not marked or
-evacuated), the free routine is called and the object is deleted from
-the list.
-
-
-The remaining objects types are all administrative --- none of them may be entered.
-
-\subsection{Thread State Objects (TSOs)}\label{sect:TSO}
-
-In the multi-threaded system, the state of a suspended thread is
-packed up into a Thread State Object (TSO) which contains all the
-information needed to restart the thread and for the garbage collector
-to find all reachable objects. When a thread is running, it may be
-``unpacked'' into machine registers and various other memory locations
-to provide faster access.
-
-Single-threaded systems don't really {\em need\/} TSOs --- but they do
-need some way to tell the storage manager about live roots so it is
-convenient to use a single TSO to store the mutator state even in
-single-threaded systems.
-
-Rather than manage TSOs' alloc/dealloc, etc., in some {\em ad hoc}
-way, we instead alloc/dealloc/etc them in the heap; then we can use
-all the standard garbage-collection/fetching/flushing/etc machinery on
-them. So that's why TSOs are ``heap objects,'' albeit very special
-ones.
-\begin{center}
-\begin{tabular}{|l|l|}
- \hline {\em Fixed header}
-\\ \hline @TSO_LINK@
-\\ \hline @TSO_WHATNEXT@
-\\ \hline @TSO_WHATNEXT_INFO@
-\\ \hline @TSO_STACK@
-\\ \hline {\em Exception Handlers}
-\\ \hline {\em Ticky Info}
-\\ \hline {\em Profiling Info}
-\\ \hline {\em Parallel Info}
-\\ \hline {\em GranSim Info}
-\\ \hline
-\end{tabular}
-\end{center}
-The contents of a TSO are:
-\begin{itemize}
-
-\item A pointer (@TSO_LINK@) used to maintain a list of threads with a similar
- state (e.g.~all runnable, all sleeping, all blocked on the same black
- hole, all blocked on the same MVar, etc.)
-
-\item A word (@TSO_WHATNEXT@) which is in suspended threads to record
- how to awaken it. This typically requires a program counter which is stored
- in the pointer @TSO_WHATNEXT_INFO@
-
-\item A pointer (@TSO_STACK@) to the top stack block.
-
-\item Optional information for ``Ticky Ticky'' statistics: @TSO_STK_HWM@ is
- the maximum number of words allocated to this thread.
-
-\item Optional information for profiling:
- @TSO_CCC@ is the current cost centre.
-
-\item Optional information for parallel execution:
-\begin{itemize}
-
-\item The types of threads (@TSO_TYPE@):
-\begin{description}
-\item[@T_MAIN@] Must be executed locally.
-\item[@T_REQUIRED@] A required thread -- may be exported.
-\item[@T_ADVISORY@] An advisory thread -- may be exported.
-\item[@T_FAIL@] A failure thread -- may be exported.
-\end{description}
-
-\item I've no idea what else
-
-\end{itemize}
-
-\item Optional information for GranSim execution:
-\begin{itemize}
-\item locked
-\item sparkname
-\item started at
-\item exported
-\item basic blocks
-\item allocs
-\item exectime
-\item fetchtime
-\item fetchcount
-\item blocktime
-\item blockcount
-\item global sparks
-\item local sparks
-\item queue
-\item priority
-\item clock (gransim light only)
-\end{itemize}
-
-
-Here are the various queues for GrAnSim-type events.
-@
-Q_RUNNING
-Q_RUNNABLE
-Q_BLOCKED
-Q_FETCHING
-Q_MIGRATING
-@
-
-\end{itemize}
-
-\subsection{Other weird objects}
-\label{sect:SPARK}
-\label{sect:BLOCKED_FETCH}
-
-\begin{description}
-\item[@BlockedFetch@ heap objects (`closures')] (parallel only)
-
-@BlockedFetch@s are inbound fetch messages blocked on local closures.
-They arise as entries in a local blocking queue when a fetch has been
-received for a local black hole. When awakened, we look at their
-contents to figure out where to send a resume.
-
-A @BlockedFetch@ closure has the form:
-\begin{center}
-\begin{tabular}{|l|l|l|l|l|l|}\hline
-{\em Fixed header} & link & node & gtid & slot & weight \\ \hline
-\end{tabular}
-\end{center}
-
-\item[Spark Closures] (parallel only)
-
-Spark closures are used to link together all closures in the spark pool. When
-the current processor is idle, it may choose to speculatively evaluate some of
-the closures in the pool. It may also choose to delete sparks from the pool.
-\begin{center}
-\begin{tabular}{|l|l|l|l|l|l|}\hline
-{\em Fixed header} & {\em Spark pool link} & {\em Sparked closure} \\ \hline
-\end{tabular}
-\end{center}
-
-
-\end{description}
-
-
-\subsection{Stack Objects}
-\label{sect:STACK_OBJECT}
-\label{sect:stacks}
-
-These are ``stack objects,'' which are used in the threaded world as
-the stack for each thread is allocated from the heap in smallish
-chunks. (The stack in the sequential world is allocated outside of
-the heap.)
-
-Each reduction thread has to have its own stack space. As there may
-be many such threads, and as any given one may need quite a big stack,
-a naive give-'em-a-big-stack-and-let-'em-run approach will cost a {\em
-lot} of memory.
-
-Our approach is to give a thread a small stack space, and then link
-on/off extra ``chunks'' as the need arises. Again, this is a
-storage-management problem, and, yet again, we choose to graft the
-whole business onto the existing heap-management machinery. So stack
-objects will live in the heap, be garbage collected, etc., etc..
-
-A stack object is laid out like this:
-
-\begin{center}
-\begin{tabular}{|l|}
-\hline
-{\em Fixed header}
-\\ \hline
-{\em Link to next stack object (0 for last)}
-\\ \hline
-{\em N, the payload size in words}
-\\ \hline
-{\em @Sp@ (byte offset from head of object)}
-\\ \hline
-{\em @Su@ (byte offset from head of object)}
-\\ \hline
-{\em Payload (N words)}
-\\ \hline
-\end{tabular}
-\end{center}
-
-\ToDo{Are stack objects on the mutable list?}
-
-The stack grows downwards, towards decreasing
-addresses. This makes it easier to print out the stack
-when debugging, and it means that a return address is
-at the lowest address of the chunk of stack it ``knows about''
-just like an info pointer on a closure.
-
-The garbage collector needs to be able to find all the
-pointers in a stack. How does it do this?
-
-\begin{itemize}
-
-\item Within the stack there are return addresses, pushed
-by @case@ expressions. Below a return address (i.e. at higher
-memory addresses, since the stack grows downwards) is a chunk
-of stack that the return address ``knows about'', namely the
-activation record of the currently running function.
-
-\item Below each such activation record is a {\em pending-argument
-section}, a chunk of
-zero or more words that are the arguments to which the result
-of the function should be applied. The return address does not
-statically
-``know'' how many pending arguments there are, or their types.
-(For example, the function might return a result of type $\alpha$.)
-
-\item Below each pending-argument section is another return address,
-and so on. Actually, there might be an update frame instead, but we
-can consider update frames as a special case of a return address with
-a well-defined activation record.
-
-\ToDo{Doesn't it {\em have} to be an update frame? After all, the arg
-satisfaction check is @Su - Sp >= ...@.}
-
-\end{itemize}
-
-The game plan is this. The garbage collector
-walks the stack from the top, traversing pending-argument sections and
-activation records alternately. Next we discuss how it finds
-the pointers in each of these two stack regions.
-
-
-\subsubsection{Activation records}\label{sect:activation-records}
-
-An {\em activation record} is a contiguous chunk of stack,
-with a return address as its first word, followed by as many
-data words as the return address ``knows about''. The return
-address is actually a fully-fledged info pointer. It points
-to an info table, replete with:
-
-\begin{itemize}
-\item entry code (i.e. the code to return to).
-\item @INFO_TYPE@ is either @RET_SMALL/RET_VEC_SMALL@ or @RET_BIG/RET_VEC_BIG@, depending
-on whether the activation record has more than 32 data words (\note{64 for 8-byte-word architectures}) and on whether
-to use a direct or a vectored return.
-\item @INFO_SM@ for @RET_SMALL@ is a bitmap telling the layout
-of the activation record, one bit per word. The least-significant bit
-describes the first data word of the record (adjacent to the fixed
-header) and so on. A ``@1@'' indicates a non-pointer, a ``@0@''
-indicates
-a pointer. We don't need to indicate exactly how many words there
-are,
-because when we get to all zeros we can treat the rest of the
-activation record as part of the next pending-argument region.
-
-For @RET_BIG@ the @INFO_SM@ field points to a block of bitmap
-words, starting with a word that tells how many words are in
-the block.
-
-\item @INFO_SRT@ is the Static Reference Table for the return
-address (Section~\ref{sect:srt}).
-\end{itemize}
-
-The activation record is a fully fledged closure too.
-As well as an info pointer, it has all the other attributes of
-a fixed header (Section~\ref{sect:fixed-header}) including a saved cost
-centre which is reloaded when the return address is entered.
-
-In other words, all the attributes of closures are needed for
-activation records, so it's very convenient to make them look alike.
-
-
-\subsubsection{Pending arguments}
-
-So that the garbage collector can correctly identify pointers
-in pending-argument sections we explicitly tag all non-pointers.
-Every non-pointer in a pending-argument section is preceded
-(at the next lower memory word) by a one-word byte count that
-says how many bytes to skip over (excluding the tag word).
-
-The garbage collector traverses a pending argument section from
-the top (i.e. lowest memory address). It looks at each word in turn:
-
-\begin{itemize}
-\item If it is less than or equal to a small constant @MAX_STACK_TAG@
-then
-it treats it as a tag heralding zero or more words of non-pointers,
-so it just skips over them.
-
-\item If it points to the code segment, it must be a return
-address, so we have come to the end of the pending-argument section.
-
-\item Otherwise it must be a bona fide heap pointer.
-\end{itemize}
-
-
-\subsection{The Stable Pointer Table}\label{sect:STABLEPTR_TABLE}
-
-A stable pointer is a name for a Haskell object which can be passed to
-the external world. It is ``stable'' in the sense that the name does
-not change when the Haskell garbage collector runs---in contrast to
-the address of the object which may well change.
-
-A stable pointer is represented by an index into the
-@StablePointerTable@. The Haskell garbage collector treats the
-@StablePointerTable@ as a source of roots for GC.
-
-In order to provide efficient access to stable pointers and to be able
-to cope with any number of stable pointers (eg $0 \ldots 100000$), the
-table of stable pointers is an array stored on the heap and can grow
-when it overflows. (Since we cannot compact the table by moving
-stable pointers about, it seems unlikely that a half-empty table can
-be reduced in size---this could be fixed if necessary by using a
-hash table of some sort.)
-
-In general a stable pointer table closure looks like this:
-
-\begin{center}
-\begin{tabular}{|l|l|l|l|l|l|l|l|l|l|l|}
-\hline
-{\em Fixed header} & {\em No of pointers} & {\em Free} & $SP_0$ & \ldots & $SP_{n-1}$
-\\\hline
-\end{tabular}
-\end{center}
-
-The fields are:
-\begin{description}
-
-\item[@NPtrs@:] number of (stable) pointers.
-
-\item[@Free@:] the byte offset (from the first byte of the object) of the first free stable pointer.
-
-\item[$SP_i$:] A stable pointer slot. If this entry is in use, it is
-an ``unstable'' pointer to a closure. If this entry is not in use, it
-is a byte offset of the next free stable pointer slot.
-
-\end{description}
-
-When a stable pointer table is evacuated
-\begin{enumerate}
-\item the free list entries are all set to @NULL@ so that the evacuation
- code knows they're not pointers;
-
-\item The stable pointer slots are scanned linearly: non-@NULL@ slots
-are evacuated and @NULL@-values are chained together to form a new free list.
-\end{enumerate}
-
-There's no need to link the stable pointer table onto the mutable
-list because we always treat it as a root.
-
-\fi
-
-\section{The Storage Manager}
-
-The generational collector remembers the depth of the last generation
-collected and the value of the heap pointer at the end of the last GC.
-If the mutator has not moved the heap pointer, that means that the
-amount of space recovered is insufficient to satisfy even one request
-and it is time to collect an older generation or report a heap overflow.
-
-A deeper collection is also triggered when a minor collection fails to
-recover at least @...@ bytes of space.
-
-When can a GC happen?
-
-@
-- During updates (ie during returns)
-- When a heap check fails
-- When a stack check fails (concurrent system only)
-- When a context switch happens (concurrent system only)
-
-When do heap checks occur?
-- Immediately after entering a thunk
-- Immediately after entering a case alternative
-
-When do stack checks occur?
-- We calculate the worst-case stack usage of an entire
- thunk so there's no need to put a check inside each alternative.
-- Immediately after entering a thunk
- We can't make a similar worst-case calculation for heap usage
- because the heap isn't used in a stacklike manner so any
- evaluation inside a case might steal some of the heap we've
- checked for.
-
-Concurrency
-- Threads can be blocked
-- Threads can be put to sleep
- - Heap may move while we sleep
- - Black holing may happen while we sleep (ie during GC)
-@
-
-\subsection{The SM state}
-
-Contains @Hp@, @HpLim@, @StablePtrTable@ plus version-specific info.
-
-\begin{itemize}
-
-\item Static Object list
-\item Foreign Object list
-\item Stable Pointer Table
-
-\end{itemize}
-
-In addition, the generational collector requires:
-
-\begin{itemize}
-
-\item Old Generation Indirection list
-\item Mutables list --- list of mutable objects in the old generation.
-\item @OldLim@ --- the boundary between the generations
-\item Old Foreign Object list --- foreign objects in the old generation
-
-\end{itemize}
-
-It is passed a table of {\em roots\/} containing
-
-\begin{itemize}
-
-\item All runnable TSOs
-
-\end{itemize}
-
-
-In the parallel system, there must be some extra magic associated with
-global GC.
-
-\subsection{The SM interface}
-
-@initSM@ finalizes any runtime parameters of the storage manager.
-
-@exitSM@ does any cleaning up required by the storage manager before
-the program is executed. Its main purpose is to print any summary
-statistics.
-
-@initHeap@ allocates the heap. It initialises the @hp@ and @hplim@
-fields of @sm@ to represent an empty heap for the compiled-in garbage
-collector. It also initialises @CAFlist@ to be the empty list. If we
-are using Appel's collector it also initialises the @OldLim@ field.
-It also initialises the stable pointer table and the @ForeignObjList@
-(and @OldForeignObjList@) fields.
-
-@collectHeap@ invokes the garbage collector. @collectHeap@ requires
-all the fields of @sm@ to be initialised appropriately (from the
-STG-machine registers). The following are identified as heap roots:
-\begin{itemize}
-\item The updated CAFs recorded in @CAFlist@.
-\item Any pointers found on the stack.
-\item All runnable and sleeping TSOs.
-\item The stable pointer table.
-\end{itemize}
-
-There are two possible results from a garbage collection:
-\begin{description}
-\item[@GC_FAIL@]
-The garbage collector is unable to free up any more space.
-
-\item[@GC_SUCCESS@]
-The garbage collector managed to free up more space.
-
-\begin{itemize}
-\item @hp@ and @hplim@ will indicate the new space available for
-allocation.
-
-\item The elements of @CAFlist@ and the stable pointers will be
-updated to point to the new locations of the closures they reference.
-
-\item Any members of @ForeignObjList@ which became garbage should have
-been reported (by calling their finalising routines; and the
-@(Old)ForeignObjList@ updated to contain only those Foreign objects
-which are still live.
-
-\end{itemize}
-
-\end{description}
-
-%************************************************************************
-%* *
-\subsection{``What really happens in a garbage collection?''}
-%* *
-%************************************************************************
-
-\ToDo{I commented out this long, out of date section - ADR}
-
-\iffalse
-
-This is a brief tutorial on ``what really happens'' going to/from the
-storage manager in a garbage collection.
-
-\begin{description}
-%------------------------------------------------------------------------
-\item[The heap check:]
-
-[OLD-ISH: WDP]
-
-If you gaze into the C output of GHC, you see many macros calls like:
-\begin{verbatim}
-HEAP_CHK_2PtrsLive((_FHS+2));
-\end{verbatim}
-
-This expands into the C (roughly speaking...):
-@
-Hp = Hp + (_FHS+2); /* optimistically move heap pointer forward */
-
-GC_WHILE_OR_IF (HEAP_OVERFLOW_OP(Hp, HpLim) OR_INTERVAL_EXPIRED) {
- STGCALL2_GC(PerformGC, <liveness-bits>, (_FHS+2));
-}
-@
-
-In the parallel world, where we will need to re-try the heap check,
-@GC_WHILE_OR_IF@ will be a ``while''; in the sequential world, it will
-be an ``if''.
-
-The ``heap lookahead'' checks, which are similar and used for
-multi-precision @Integer@ ops, have some further complications. See
-the commentary there (@StgMacros.lh@).
-
-%------------------------------------------------------------------------
-\item[Into @callWrapper_GC@...:]
-
-When we failed the heap check (above), we were inside the
-GCC-registerised ``threaded world.'' @callWrapper_GC@ is all about
-getting in and out of the threaded world. On SPARCs, with register
-windows, the name of the game is not shifting windows until we have
-what we want out of the old one. In tricky cases like this, it's best
-written in assembly language.
-
-Performing a GC (potentially) means giving up the thread of control.
-So we must fill in the thread-state-object (TSO) [and its associated
-stk object] with enough information for later resumption:
-\begin{enumerate}
-\item
-Save the return address in the TSO's PC field.
-\item
-Save the machine registers used in the STG threaded world in their
-corresponding TSO fields. We also save the pointer-liveness
-information in the TSO.
-\item
-The registers that are not thread-specific, notably @Hp@ and
-@HpLim@, are saved in the @StorageMgrInfo@ structure.
-\item
-Call the routine it was asked to call; in this example, call
-@PerformGC@ with arguments @<liveness>@ and @_FHS+2@ (some constant)...
-
-\end{enumerate}
-
-%------------------------------------------------------------------------
-\item[Into the heap overflow wrapper, @PerformGC@ [parallel]:]
-
-Most information has already been saved in the TSO.
-
-\begin{enumerate}
-\item
-The first argument (@<liveness>@, in our example) say what registers
-are live, i.e., are ``roots'' the storage manager needs to know.
-\begin{verbatim}
-StorageMgrInfo.rootno = 2;
-StorageMgrInfo.roots[0] = (P_) Ret1_SAVE;
-StorageMgrInfo.roots[1] = (P_) Ret2_SAVE;
-\end{verbatim}
-
-\item
-We move the heap-pointer back [we had optimistically
-advanced it, in the initial heap check]
-
-\item
-We load up the @smInfo@ data from the STG registers' @*_SAVE@ locations.
-
-\item
-We mark on the scheduler's big ``blackboard'' that a GC is
-required.
-
-\item
-We reschedule, i.e., this thread gives up control. (The scheduler
-will presumably initiate a garbage-collection, but it may have to do
-any number of other things---flushing, for example---before ``normal
-execution'' resumes; and it most certainly may not be this thread that
-resumes at that point!)
-\end{enumerate}
-
-IT IS AT THIS POINT THAT THE WORLD IS COMPLETELY TIDY.
-
-%------------------------------------------------------------------------
-\item[Out of @callWrapper_GC@ [parallel]:]
-
-When this thread is finally resumed after GC (and who knows what
-else), it will restart by the normal enter-TSO/enter-stack-object
-sequence, which has the effect of re-loading the registers, etc.,
-(i.e., restoring the state).
-
-Because the address we saved in the TSO's PC field was that at the end
-of the heap check, and because the check is a while-loop in the
-parallel system, we will now loop back around, and make sure there is
-enough space before continuing.
-\end{description}
-
-\fi % end of commented out part
-
-\subsection{Static Reference Tables (SRTs)}
-\label{sect:srt}
-\label{sect:CAF}
-\label{sect:static-objects}
-
-In the above, we assumed that objects always contained pointers to all
-their free variables. In fact, this isn't quite true: GHC omits
-pointers to top-level objects and allocates their closures in static
-memory. This optimisation reduces the number of free variables in
-heap objects - reducing memory usage and the effort needed to put them
-into heap objects. However, this optimisation comes at a cost: we
-need to complicate the garbage collector with machinery for tracing
-these static references.
-
-Early versions of GHC used a very simple algorithm: it treated all
-static objects as roots. This is safe in the sense that no object is
-ever deallocated if there's a chance that it might be required later
-but can lead to some terrible space leaks. For example, this program
-ought to be able to run in constant space but, because @xs@ is never
-deallocated, it runs in linear space.
-
-@
-main = print xs
-xs = [1..]
-@
-
-The correct behaviour is for the garbage collector to keep a static
-object alive iff it might be required later in execution. That is, if
-it is reachable from any live heap objects {\em or\/} from any return
-addresses found on the stack or from the current program counter.
-Since it is obviously infeasible for the garbage collector to scan
-machine code looking for static references, the code generator must
-generate a table of all static references in any piece of code (and we
-must place a pointer to this table next to any piece of code we
-generate).
-
-Here's what the SRT has to contain:
-
-@
-...
-@
-
-Here's how we represent it:
-
-@
-...
-must be able to handle 0 references well
-@
-
-@
-Other trickery:
-o The CAF list
-o The scavenge list
-o Generational GC trickery
-@
-
-\subsection{Space leaks and black holes}
-\label{sect:black-hole}
-
-\iffalse
-
-\ToDo{Insert text stolen from update paper}
-
-\else
-
-A program exhibits a {\em space leak} if it retains storage that is
-sure not to be used again. Space leaks are becoming increasingly
-common in imperative programs that @malloc@ storage and fail
-subsequently to @free@ it. They are, however, also common in
-garbage-collected systems, especially where lazy evaluation is
-used.[.wadler leak, runciman heap profiling jfp.]
-
-Quite a bit of experience has now accumulated suggesting that
-implementors must be very conscientious about avoiding gratuitous
-space leaks --- that is, ones which are an accidental artefact of some
-implementation technique.[.appel book.] The update mechanism is
-a case in point, as <.jones jfp leak.> points out. Consider a thunk for
-the expression
-@
- let xs = [1..1000] in last xs
-@
-where @last@ is a function that returns the last element of its
-argument list. When the thunk is entered it will call @last@, which
-will consume @xs@ until it finds the last element. Since the list
-@[1..1000]@ is produced lazily one might reasonably expect the
-expression to evaluate in constant space. But {\em until the moment
-of update, the thunk itself still retains a pointer to the beginning
-of the list @xs@}. So, until the update takes place the whole list
-will be retained!
-
-Of course, this is completely gratuitous. The pointer to @xs@ in the
-thunk will never be used again. In <.peyton stock hardware.> the solution to
-this problem that we advocated was to overwrite a thunk's info with a
-fixed ``black hole'' info pointer, {\em at the moment of entry}. The
-storage management information attached to a black-hole info pointer
-tells the garbage collector that the closure contains no pointers,
-thereby plugging the space leak.
-
-\subsubsection{Lazy black-holing}
-\label{sect:lazy-black-holing}
-
-\Note{We currently plan to implement eager black holing because the
-lazy blackholing scheme leavs "slop" in the heap.}
-
-Black-holing is a well-known idea. The trouble is that it is
-gallingly expensive. To avoid the occasional space leak, for every
-single thunk entry we have to load a full-word literal constant into a
-register (often two instructions) and then store that register into a
-memory location.
-
-Fortunately, this cost can easily be avoided. The
-idea is simple: {\em instead of black-holing every thunk on entry,
-wait until the garbage collector is called, and then black-hole all
-(and only) the thunks whose evaluation is in progress at that moment}.
-There is no benefit in black-holing a thunk that is updated before
-garbage collection strikes! In effect, the idea is to perform the
-black-holing operation lazily, only when it is needed. This
-dramatically cuts down the number of black-holing operations, as our
-results show {\em forward ref}.
-
-How can we find all the thunks whose evaluation is in progress? They
-are precisely the ones for which update frames are on the stack. So
-all we need do is find all the update frames (via the @Su@ chain) and
-black-hole their thunks right at the start of garbage collection.
-Notice that it is not enough to refrain from treating update frames as
-roots: firstly because the thunks to which they point may need to be
-moved in a copying collector, but more importantly because the thunk
-might be accessible via some other route.
-
-\subsubsection{Detecting loops}
-
-Black-holing has a second minor advantage: evaluation of a thunk whose
-value depends on itself will cause a black hole closure to be entered,
-which can cause a suitable error message to be displayed. For example,
-consider the definition
-@
- x = 1+x
-@
-The code to evaluate @x@'s right hand side will evaluate @x@. In the
-absence of black-holing, the result will be a stack overflow, as the
-evaluator repeatedly pushes a return address and enters @x@. If
-thunks are black-holed on entry, then this infinite loop can be caught
-almost instantly.
-
-With our new method of lazy black-holing, a self-referential program
-might cause either stack overflow or a black-hole error message,
-depending on exactly when garbage collection strikes. It is quite
-easy to conceal these differences, however. If stack overflow occurs,
-all we need do is examine the update frames on the stack to see if
-more than one refers to the same thunk. If so, there is a loop that
-would have been detected by eager black-holing.
-
-\subsubsection{Lazy locking}
-\label{sect:lock}
-
-In a parallel implementation, it is necessary somehow to ``lock'' a
-thunk that is under evaluation, so that other parallel evaluators
-cannot simultaneously evaluate it and thereby duplicate work.
-Instead, an evaluator that enters a locked thunk should be blocked,
-and made runnable again when the thunk is updated.
-
-This locking is readily arranged in the same way as black-holing, by
-overwriting the thunk's info pointer with a special ``locked'' info
-pointer, at the moment of entry. If another evaluator enters the
-thunk before it has been updated, it will land in the entry code for
-the ``locked'' info pointer, which blocks the evaluator and queues it
-on the locked thunk.
-
-The details are given by <.portable parallel trinder.>. However, the close similarity
-between locking and black holing suggests the following question: can
-locking be done lazily too? The answer is that it can, except that
-locking can be postponed only until the next {\em context switch},
-rather than the next {\em garbage collection}. We are assuming here
-that the parallel implementation does not use shared memory to allow
-two processors to access the same closure. If such access is
-permitted then every thunk entry requires a hardware lock, and becomes
-much too expensive.
-
-Is lazy locking worth while, given that it requires extra work every
-context switch? We believe it is, because contexts switches are
-relatively infrequent, and thousands of thunk-entries typically take
-place between each.
-
-{\em Measurements elsewhere. Omit this section? If so, fix cross refs to here.}
-
-\fi
-
-
-\subsection{Squeezing identical updates}
-
-\Note{This can also be done by testing whether @Sp == Su@ when we push
-an update frame. If so, we can overwrite the updatee with an
-indirection to the existing updatee (and some slop objects) and avoid
-pushing an update frame.}
-
-\iffalse
-
-\ToDo{Insert text stolen from update paper}
-
-\else
-
-Consider the following Haskell definition of the standard
-function @partition@ that divides a list into two, those elements
-that satisfy a predicate @p@ and those that do not:
-@
- partition :: (a->Bool) -> [a] -> ([a],[a])
- partition p [] = ([],[])
- partition p (x:xs) = if p x then (x:ys, zs)
- else (ys, x:zs)
- where
- (ys,zs) = partition p xs
-@
-By the time this definition has been desugared, it looks like this:
-@
- partition p xs
- = case xs of
- [] -> ([],[])
- (x:xs) -> let
- t = partition p xs
- ys = fst t
- zs = snd t
- in
- if p x then (x:ys,zs)
- else (ys,x:zs)
-@
-Lazy evaluation demands that the recursive call is bound to an
-intermediate variable, @t@, from which @ys@ and @zs@ are lazily
-selected. (The functions @fst@ and @snd@ select the first and second
-elements of a pair, respectively.)
-
-Now, suppose that @partition@ is applied to a list @[x1,x2]@,
-all of whose
-elements satisfy @p@. We can get a good idea of what will happen
-at runtime by unrolling the recursion a few times in our heads.
-Unrolling once, and remembering that @(p x1)@ is @True@, we get this:
-@
- partition p [x1,x2]
-=
- let t1 = partition [x2]
- ys1 = fst t1
- zs1 = snd t1
- in (x1:ys1, zs1)
-@
-Unrolling the rest of the way gives this:
-@
- partition p [x1,x2]
-=
- let t2 = ([],[])
- ys2 = fst t2
- zs2 = snd t2
- t1 = (x2:ys2,zs2)
- ys1 = fst t1
- zs1 = snd t1
- in (x1:ys1,zs1)
-@
-Now consider what happens if @zs1@ is evaluated. It is bound to a
-thunk, which will push an update frame before evaluating the
-expression @snd t1@. This expression in turn forces evaluation of
-@zs2@, which pushes an update frame before evaluating @snd t2@.
-Indeed the stack of update frames will grow as deep as the list is
-long when @zs1@ is evaluated. This is rather galling, since all the
-thunks @zs1@, @zs2@, and so on, have the same value.
-
-\ToDo{Describe the state-transformer case in which we get a space leak from
-pending update frames.}
-
-The solution is simple. The garbage collector, which is going to traverse the
-update stack in any case, can easily identify two update frames that are directly
-on top of each other. The second of these will update its target with the same
-value as the first. Therefore, the garbage collector can perform the update
-right away, by overwriting one update target with an indirection to the second,
-and eliminate the corresponding update frame. In this way ever-growing stacks of
-update frames are reduced to a single representative at garbage collection time.
-If this is done at the start of garbage collection then, if it turns out that
-some of these update targets are garbage they will be collected right away.
-
-\fi
-
-\subsection{Space leaks and selectors}\label{sect:space-leaks-and-selectors}
-
-\iffalse
-
-\ToDo{Insert text stolen from update paper}
-
-\else
-
-In 1987, Wadler identified an important source of space leaks in
-lazy functional programs. Consider the Haskell function definition:
-@
- f p = (g1 a, g2 b) where (a,b) = p
-@
-The pattern-matching in the @where@ clause is known as
-{\em lazy pattern-matching}, because it is performed only if @a@
-or @b@ is actually evaluated. The desugarer translates lazy pattern matching
-to the use of selectors, @fst@ and @snd@ in this case:
-@
- f p = let a = fst p
- b = snd p
- in
- (b, a)
-@
-Now suppose that the second component of the pair @(f p)@, namely @a@,
-is evaluated and discarded, but the first is not although it remains
-reachable. The garbage collector will find that the thunk for @b@ refers
-to @p@ and hence to @a@. Thus, although @a@ cannot ever be used again, its
-space is retained. It turns out that this space leak can have a very bad effect
-indeed on a program's space behaviour (Section~\ref{sect:selector-results}).
-
-Wadler's paper also proposed a solution: if the garbage collector
-encounters a thunk of the form @snd p@, where @p@ is evaluated, then
-the garbage collector should perform the selection and overwrite the
-thunk with a pointer to the second component of the pair. In effect, the
-garbage collector thereby performs a bounded amount of as-yet-undemanded evaluation
-in the hope of improving space behaviour.
-We implement this idea directly, by making the garbage collector
-eagerly execute all selector thunks\footnote{A word of caution: it is rather easy
-to make a mistake in the implementation, especially if the garbage collector
-uses pointer reversal to traverse the reachable graph.},
-with results
-reported in Section~\ref{sect:THUNK_SEL}.
-
-One could easily imagine generalisations of this idea, with the garbage
-collector performing bounded amounts of space-saving work. One example is
-this:
-@
- f x [] = (x,x)
- f x (y:ys) = f (x+1) ys
-@
-Most lazy evaluators will build up a chain of thunks for the accumulating
-parameter, @x@, each of which increments @x@. It is not safe to evaluate
-any of these thunks eagerly, since @f@ is not strict in @x@, and we know nothing
-about the value of @x@ passed in the initial call to @f@.
-On the other hand, if the garbage collector found a thunk @(x+1)@ where
-@x@ happened to be evaluated, then it could ``execute'' it eagerly.
-If done carefully, the entire chain could be eliminated in a single
-garbage collection. We have not (yet) implemented this idea.
-A very similar idea, dubbed ``stingy evaluation'', is described
-by <.stingy.>.
-
-\ToDo{Simple generalisation: handle all the ``standard closures'' this way.}
-
-<.sparud lazy pattern matching.> describes another solution to the
-lazy-pattern-matching
-problem. His solution involves adding code to the two thunks for
-@a@ and @b@ so that if either is evaluated it arranges to update the
-other as well as itself. The garbage-collector solution is a little
-more general, since it applies whether or not the selectors were
-generated by lazy pattern matching, and in our setting it was easier
-to implement than Sparud's.
-
-\fi
-
-
-\subsection{Internal workings of the Compacting Collector}
-
-\subsection{Internal workings of the Copying Collector}
-
-\subsection{Internal workings of the Generational Collector}
-
-
-\section{Profiling}
-
-Registering costs centres looks awkward - can we tidy it up?
-
-\section{Parallelism}
-
-Something about global GC, inter-process messages and fetchmes.
-
-\section{Debugging}
-
-\section{Ticky Ticky profiling}
-
-Measure what proportion of ...:
-\begin{itemize}
-\item
-... Enters are to data values, function values, thunks.
-\item
-... allocations are for data values, functions values, thunks.
-\item
-... updates are for data values, function values.
-\item
-... updates ``fit''
-\item
-... return-in-heap (dynamic)
-\item
-... vectored return (dynamic)
-\item
-... updates are wasted (never re-entered).
-\item
-... constructor returns get away without hitting an update.
-\end{itemize}
-
-%************************************************************************
-%* *
-\subsection[ticky-stk-heap-use]{Stack and heap usage}
-%* *
-%************************************************************************
-
-Things we are interested in here:
-\begin{itemize}
-\item
-How many times we do a heap check and move @Hp@; comparing this with
-the allocations gives an indication of how many things we get per trip
-to the well:
-
-If we do a ``heap lookahead,'' we haven't really allocated any
-heap, so we need to undo the effects of an @ALLOC_HEAP@:
-
-\item
-The stack high-water mark.
-
-\item
-Re-use of stack slots, and stubbing of stack slots:
-
-\end{itemize}
-
-%************************************************************************
-%* *
-\subsection[ticky-allocs]{Allocations}
-%* *
-%************************************************************************
-
-We count things every time we allocate something in the dynamic heap.
-For each, we count the number of words of (1)~``admin'' (header),
-(2)~good stuff (useful pointers and data), and (3)~``slop'' (extra
-space, in hopes it will allow an in-place update).
-
-The first five macros are inserted when the compiler generates code
-to allocate something; the categories correspond to the @ClosureClass@
-datatype (manifest functions, thunks, constructors, big tuples, and
-partial applications).
-
-We may also allocate space when we do an update, and there isn't
-enough space. These macros suffice (for: updating with a partial
-application and a constructor):
-
-In the threaded world, we allocate space for the spark pool, stack objects,
-and thread state objects.
-
-The histogrammy bit is fairly straightforward; the @-2@ is: one for
-0-origin C arrays; the other one because we do {\em no} one-word
-allocations, so we would never inc that histogram slot; so we shift
-everything over by one.
-
-Some hard-to-account-for words are allocated by/for primitives,
-includes Integer support. @ALLOC_PRIM2@ tells us about these. We
-count everything as ``goods'', which is not strictly correct.
-(@ALLOC_PRIM@ is the same sort of stuff, but we know the
-admin/goods/slop breakdown.)
-
-%************************************************************************
-%* *
-\subsection[ticky-enters]{Enters}
-%* *
-%************************************************************************
-
-We do more magical things with @ENT_FUN_DIRECT@. Besides simply knowing
-how many ``fast-entry-point'' enters there were, we'd like {\em simple}
-information about where those enters were, and the properties thereof.
-@
-struct ent_counter {
- unsigned registeredp:16, /* 0 == no, 1 == yes */
- arity:16, /* arity (static info) */
- Astk_args:16, /* # of args off A stack */
- Bstk_args:16; /* # of args off B stack */
- /* (rest of args are in registers) */
- StgChar *f_str; /* name of the thing */
- StgChar *f_arg_kinds; /* info about the args types */
- StgChar *wrap_str; /* name of its wrapper (if any) */
- StgChar *wrap_arg_kinds;/* info about the orig wrapper's arg types */
- I_ ctr; /* the actual counter */
- struct ent_counter *link; /* link to chain them all together */
-};
-@
-
-%************************************************************************
-%* *
-\subsection[ticky-returns]{Returns}
-%* *
-%************************************************************************
-
-Whenever a ``return'' occurs, it is returning the constituent parts of
-a data constructor. The parts can be returned either in registers, or
-by allocating some heap to put it in (the @ALLOC_*@ macros account for
-the allocation). The constructor can either be an existing one
-(@*OLD*@) or we could have {\em just} figured out this stuff
-(@*NEW*@).
-
-Here's some special magic that Simon wants [edited to match names
-actually used]:
-
-@
-From: Simon L Peyton Jones <simonpj>
-To: partain, simonpj
-Subject: counting updates
-Date: Wed, 25 Mar 92 08:39:48 +0000
-
-I'd like to count how many times we update in place when actually Node
-points to the thing. Here's how:
-
-@RET_OLD_IN_REGS@ sets the variable @ReturnInRegsNodeValid@ to @True@;
-@RET_NEW_IN_REGS@ sets it to @False@.
-
-@RET_SEMI_???@ sets it to??? ToDo [WDP]
-
-@UPD_CON_IN_PLACE@ tests the variable, and increments @UPD_IN_PLACE_COPY_ctr@
-if it is true.
-
-Then we need to report it along with the update-in-place info.
-@
-
-
-Of all the returns (sum of four categories above), how many were
-vectored? (The rest were obviously unvectored).
-
-%************************************************************************
-%* *
-\subsection[ticky-update-frames]{Update frames}
-%* *
-%************************************************************************
-
-These macros count up the following update information.
-
-\begin{tabular}{|l|l|} \hline
-Macro & Counts \\ \hline
- & \\
-@UPDF_STD_PUSHED@ & Update frame pushed \\
-@UPDF_CON_PUSHED@ & Constructor update frame pushed \\
-@UPDF_HOLE_PUSHED@ & An update frame to update a black hole \\
-@UPDF_OMITTED@ & A thunk decided not to push an update frame \\
- & (all subsets of @ENT_THK@) \\
-@UPDF_RCC_PUSHED@ & Cost Centre restore frame pushed \\
-@UPDF_RCC_OMITTED@ & Cost Centres not required -- not pushed \\\hline
-\end{tabular}
-
-%************************************************************************
-%* *
-\subsection[ticky-updates]{Updates}
-%* *
-%************************************************************************
-
-These macros record information when we do an update. We always
-update either with a data constructor (CON) or a partial application
-(PAP).
-
-\begin{tabular}{|l|l|}\hline
-Macro & Where \\ \hline
- & \\
-@UPD_EXISTING@ & Updating with an indirection to something \\
- & already in the heap \\
-@UPD_SQUEEZED@ & Same as @UPD_EXISTING@ but because \\
- & of stack-squeezing \\
-@UPD_CON_W_NODE@ & Updating with a CON: by indirecting to Node \\
-@UPD_CON_IN_PLACE@ & Ditto, but in place \\
-@UPD_CON_IN_NEW@ & Ditto, but allocating the object \\
-@UPD_PAP_IN_PLACE@ & Same, but updating w/ a PAP \\
-@UPD_PAP_IN_NEW@ & \\\hline
-\end{tabular}
-
-%************************************************************************
-%* *
-\subsection[ticky-selectors]{Doing selectors at GC time}
-%* *
-%************************************************************************
-
-@GC_SEL_ABANDONED@: we could've done the selection, but we gave up
-(e.g., to avoid overflowing the C stack); @GC_SEL_MINOR@: did a
-selection in a minor GC; @GC_SEL_MAJOR@: ditto, but major GC.
-
-
-
-\section{History}
-
-We're nuking the following:
-
-\begin{itemize}
-\item
- Two stacks
-
-\item
- Return in registers.
- This lets us remove update code pointers from info tables,
- removes the need for phantom info tables, simplifies
- semi-tagging, etc.
-
-\item
- Threaded GC.
- Careful analysis suggests that it doesn't buy us very much
- and it is hard to work with.
-
- Eliminating threaded GCs eliminates the desire to share SMReps
- so they are (once more) part of the Info table.
-
-\item
- RetReg.
- Doesn't buy us anything on a register-poor architecture and
- isn't so important if we have semi-tagging.
-
-@
- - Probably bad on register poor architecture
- - Can avoid need to write return address to stack on reg rich arch.
- - when a function does a small amount of work, doesn't
- enter any other thunks and then returns.
- eg entering a known constructor (but semitagging will catch this)
- - Adds complications
-@
-
-\item
- Update in place
-
- This lets us drop CONST closures and CHARLIKE closures (assuming we
- don't support Unicode). The only point of these closures was to
- avoid updating with an indirection.
-
- We also drop @MIN_UPD_SIZE@ --- all we need is space to insert an
- indirection or a black hole.
-
-\item
- STATIC SMReps are now called CONST
-
-\item
- @SM_MUTVAR@ is new
+\item
+ @MUTVAR@ is new
\item The profiling ``kind'' field is now encoded in the @INFO_TYPE@ field.
This identifies the general sort of the closure for profiling purposes.
\end{itemize}
-\section{Old tricks}
-
-@CAF@ indirections:
-
-These are statically defined closures which have been updated with a
-heap-allocated result. Initially these are exactly the same as a
-@STATIC@ closure but with special entry code. On entering the closure
-the entry code must:
-
-\begin{itemize}
-\item Allocate a black hole in the heap which will be updated with
- the result.
-\item Overwrite the static closure with a special @CAF@ indirection.
-
-\item Link the static indirection onto the list of updated @CAF@s.
-\end{itemize}
-
-The indirection and the link field require the initial @STATIC@
-closure to be of at least size @MIN_UPD_SIZE@ (excluding the fixed
-header).
-
-@CAF@s are treated as special garbage collection roots. These roots
-are explicitly collected by the garbage collector, since they may
-appear in code even if they are not linked with the main heap. They
-consequently represent potentially enormous space-leaks. A @CAF@
-closure retains a fixed location in statically allocated data space.
-When updated, the contents of the @CAF@ indirection are changed to
-reflect the new closure. @CAF@ indirections require special garbage
-collection code.
-
-\section{Old stuff about SRTs}
-
-\ToDo{Commented out}
-
-\iffalse
-
-Garbage collection of @CAF@s is tricky. We have to cope with explicit
-collection from the @CAFlist@ as well as potential references from the
-stack and heap which will cause the @CAF@ evacuation code to be
-called. They are treated like indirections which are shorted out.
-However they must also be updated to point to the new location of the
-new closure as the @CAF@ may still be used by references which
-reside in the code.
-
-{\bf Copying Collection}
-
-A first scheme might use evacuation code which evacuates the reference
-and updates the indirection. This is no good as subsequent evacuations
-will result in an already evacuated closure being evacuated. This will
-leave a forward reference in to-space!
-
-An alternative scheme evacuates the @CAFlist@ first. The closures
-referenced are evacuated and the @CAF@ indirection updated to point to
-the evacuated closure. The @CAF@ evacuation code simply returns the
-updated indirection pointer --- the pointer to the evacuated closure.
-Unfortunately the closure the @CAF@ references may be a static
-closure, in fact, it may be another @CAF@. This will cause the second
-@CAF@'s evacuation code to be called before the @CAF@ has been
-evacuated, returning an unevacuated pointer.
-
-Another scheme leaves updating the @CAF@ indirections to the end of
-the garbage collection. All the references are evacuated and
-scavenged as usual (including the @CAFlist@). Once collection is
-complete the @CAFlist@ is traversed updating the @CAF@ references with
-the result of evacuating the referenced closure again. This will
-immediately return as it must be a forward reference, a static
-closure, or a @CAF@ which will indirect by evacuating its reference.
-
-The crux of the problem is that the @CAF@ evacuation code needs to
-know if its reference has already been evacuated and updated. If not,
-then the reference can be evacuated, updated and returned safely
-(possibly evacuating another @CAF@). If it has, then the updated
-reference can be returned. This can be done using two @CAF@
-info-tables. At the start of a collection the @CAFlist@ is traversed
-and set to an internal {\em evacuate and update} info-table. During
-collection, evacution of such a @CAF@ also results in the info-table
-being reset back to the standard @CAF@ info-table. Thus subsequent
-evacuations will simply return the updated reference. On completion of
-the collection all @CAF@s will have {\em return reference} info-tables
-again.
-
-This is the scheme we adopt. A @CAF@ indirection has evacuation code
-which returns the evacuated and updated reference. During garbage
-collection, all the @CAF@s are overwritten with an internal @CAF@ info
-table which has evacuation code which performs this evacuate and
-update and restores the original @CAF@ code. At some point during the
-collection we must ensure that all the @CAF@s are indeed evacuated.
-
-The only potential problem with this scheme is a cyclic list of @CAF@s
-all directly referencing (possibly via indirections) another @CAF@!
-Evacuation of the first @CAF@ will fail in an infinite loop of @CAF@
-evacuations. This is solved by ensuring that the @CAF@ info-table is
-updated to a {\em return reference} info-table before performing the
-evacuate and update. If this {\em return reference} evacuation code is
-called before the actual evacuation is complete it must be because
-such a cycle of references exists. Returning the still unevacuated
-reference is OK --- all the @CAF@s will now reference the same
-@CAF@ which will reference itself! Construction of such a structure
-indicates the program must be in an infinite loop.
-
-{\bf Compacting Collector}
-
-When shorting out a @CAF@, its reference must be marked. A first
-attempt might explicitly mark the @CAF@s, updating the reference with
-the marked reference (possibly short circuting indirections). The
-actual @CAF@ marking code can indicate that they have already been
-marked (though this might not have actually been done yet) and return
-the indirection pointer so it is shorted out. Unfortunately the @CAF@
-reference might point to an indirection which will be subsequently
-shorted out. Rather than returning the @CAF@ reference we treat the
-@CAF@ as an indirection, calling the mark code of the reference, which
-will return the appropriately shorted reference.
-
-Problem: Cyclic list of @CAF@s all directly referencing (possibly via
-indirections) another @CAF@!
-
-Before compacting, the locations of the @CAF@ references are
-explicitly linked to the closures they reference (if they reference
-heap allocated closures) so that the compacting process will update
-them to the closure's new location. Unfortunately these locations'
-@CAF@ indirections are static. This causes premature termination
-since the test to find the info pointer at the end of the location
-list will match more than one value. This can be solved by using an
-auxiliary dynamic array (on the top of the A stack). One location for
-each @CAF@ indirection is linked to the closure that the @CAF@
-references. Once collection is complete this array is traversed and
-the corresponding @CAF@ is then updated with the updated pointer from
-the auxiliary array.
-
-
-It is possible to use an alternative marking scheme, using a similar
-idea to the copying solution. This scheme avoids the need to update
-the @CAF@ references explicitly. We introduce an auxillary {\em mark
-and update} @CAF@ info-table which is used to update all @CAF@s at the
-start of a collection. The new code marks the @CAF@ reference,
-updating it with the returned reference. The returned reference is
-itself returned so the @CAF@ is shorted out. The code also modifies the
-@CAF@ info-table to be a {\em return reference}. Subsequent attempts to
-mark the @CAF@ simply return the updated reference.
-
-A cyclic @CAF@ reference will result in an attempt to mark the @CAF@
-before the marking has been completed and the reference updated. We
-cannot start marking the @CAF@ as it is already being marked. Nor can
-we return the reference as it has not yet been updated. Neither can we
-treat the CAF as an indirection since the @CAF@ reference has been
-obscured by the pointer reversal stack. All we can do is return the
-@CAF@ itself. This will result in some @CAF@ references not being
-shorted out.
-
-This scheme has not been adopted but has been implemented. The code is
-commented out with @#if 0@.
-
-\fi
-
-\subsection{The virtual register set}
-
-\ToDo{Commented out}
-
-\iffalse
-
-We refer to any (atomic) part of the virtual machine state as a ``register.''
-These ``registers'' may be shared between all threads in the system or may be
-specific to each thread.
-
-Global:
-@
- Hp
- HpLim
- Thread preemption flag
-@
-
-Thread specific:
-@
- TSO - pointer to the TSO for when we have to pack thread away
- Sp
- SpLim
- Su - used to calculate number of arguments on stack
- - this is a more convenient representation
- Call/return registers (aka General purpose registers)
- Cost centre (and other debug/profile info)
- Statistic gathering (not in production system)
- Exception handlers
- Heap overflow - possible global?
- Stack overflow - possibly global?
- Pattern match failure
- maybe a failWith handler?
- maybe an exitWith handler?
- ...
-@
-
-Some of these virtual ``registers'' are used very frequently and should
-be mapped onto machine registers if at all possible. Others are used
-very infrequently and can be kept in memory to free up registers for
-other uses.
-
-On register-poor architectures, we can play a few tricks to reduce the
-number of virtual registers which need to be accessed on a regular
-basis:
-
-@
-- HpLim trick
-- Grow stack and heap towards each other (single-threaded system only)
-- We might need to keep the C stack pointer in a register if that
- is what the OS expects when a signal occurs.
-- Preemption flag trick
-- If any of the frequently accessed registers cannot be mapped onto
- machine registers we should keep the TSO in a machine register to
- allow faster access to all the other non-machine registers.
-@
-
-\fi
\end{document}
projects / ghc-hetmet.git / commitdiff