Trailing-Edge
-
PDP-10 Archives
-
decuslib10-05
-
43,50337/25/tdiii.rno
There is 1 other file named tdiii.rno in the archive. Click here to see a list.
.sp 1;.nonum;.lm 8;.ps 50,72
.pg;.ts 8,16,24,32,40,48,56,64,72,80
.nm ch 1
.ch SIMULA FOR DEC SYSTEM 10#############TD, RTS
.c;SIMULA 67 FOR DEC SYSTEM 10
.s 2;.c;THE RUN TIME SYSTEM
.s 2;.c;74-11-13 Lars Enderin
.nm 1
.st 741111############################################Lars Enderin
.pg;III##THE RUN TIME SYSTEM (RTS)
.br;#####-------------------------
.s;III.1 RTS DESCRIPTION
.p 0,2,10;III.1.1##Introduction
.br;#########------------
.s;The Run Time System (RTS) is a collection of subroutines designed to support
an executing SIMULA program.
The RTS consists of low segment parts loaded on demand with the compiled
program and high segment parts loaded at run time.
The debugging subsystem SIMDDT may be loaded at program start, via REENTER
after a _^C interrupt, or after program exit for a new execution.
SIMDDT is always loaded on run time errors.
.p 0,4,15;III.1.2##RTS design goals and principles
.br;#########-------------------------------
.s;The RTS is designed to take advantage of the hardware characteristics
of the KI-10 CPU. The data structures have been designed with a view to
efficient operation in a paged virtual memory environment. The number
of storage allocations has been reduced by allocating most subblocks within
already existing blocks.
Data referencing locality has been improved by allocation of display vectors
adjacent to their block instances.
Calling sequences and data representation have been designed to be efficient.
The most commonly occurring sequences have been optimized.
.pg;III.1.4##RTS coding and naming conventions
.br;#########---------------------------------
.p 0,2,10;III.1.4.1##Naming conventions in the RTS
.s;As in the compiler, a fairly rigid naming scheme has been observed to aid in
deducing the meaning of a name in the system and to simplify consistency
checks by visual inspection.
Normally, the first letter signifies the type of quantity as follows:
.s;Q starts a compile-time constant.
.s;S or SW signify a switch (Boolean variable) which is a 1-bit or 36-bit field
in an accumulator or storage cell.
.s;X stands for an accumulator.
.s;Y is a global cell or a local cell in a subroutine. Also used for offsets in
the static low segment area, based on the value in .JBOPS.
.s;YY in some subroutine specifications is used to signify a formal parameter.
.s;XX is used in some routines when designating a formal parameter in an ac.
.s;Z starts a record field designator, defined via the DF macro (see I.6).
Record names (used in DR macros and prefixed by Q to form a record type code)
are three letters starting with Z. Record field designators consist of a
record name suffixed by 2 or 3 letters and/or digits. The first 5 characters of
a field designator should be unique since symbols are derived from it by
prefixing special characters (% and $).
.s;Component names are two letters, e g CP (class and prefixed block handling).
Subroutine names are formed from component names followed by two and sometimes
more characters (two preferred). The symbols L1, L2, ... , L10 are preferrably
used as local labels (available within BEGIN - ENDD or PROC - EPROC),
otherwise the component name is used as prefix.
Global symbols (INTERN or ENTRY) always have a dot "." as prefix to avoid
name clashes with SIMULA procedures.
OPDEF's have been used in some modules as a convenient means of
coding procedure calls.
.p 0,4,6;III.1.4.2##Register conventions
.s;The accumulators (registers) have been given the basic mnemonics
X0, X1, ..., X7, X10, ..., X17 (octal numbering). These names are used
in local contexts only.
.s;X0
.br;With very few exceptions,
X0 is used only internally in the RTS, e g to hold the first word (with the
flags) of a dynamic record. One exception is the mathematical subroutines,
which return results in X0 and possibly X1 (=XSAC).
.s;X1
.br;X1 is called XSAC and is used for various purposes, e g to pass prototype
address or the like to a RTS routine. It is used as a parameter register
chiefly when no extra code has to be executed in the compiler, which
normally compiles code using XWAC1 etc.
In other words, XSAC is used when the parameter
to be passed could not be compiled to some work ac more easily.
.s;X2
.br;X2 is called XTAC and is used for example to pass the number of the top ac
to standard functions.
.s;X0 through X2 can normally be used in the RTS without being saved and restored
(except for internal subroutine interface).
.s;X3
.br;
X3 is called XWAC1 and also XRAC. This is the
first ac allocated by the compiler and also the result ac from thunks,
procedures and RTS functions, together with XWAC2 (=XRAC1) in some cases.
New object addresses are also computed to this accumulator.
.s;X4-X14
.br;
X4-X14 (octal) are the rest of the work accumulators, also called
XWAC2, XWAC3, ..., XWAC7, XWAC10, XWAC11, XWAC12. The upper limit is defined
by the compile-time constant QNAC which is the number of work ac's.
.s;X15
.br;X15 is called XCB and should always point to the nearest (current) block with a
display vector attached.
XCB is the root of the run-time data structure. From this all referenceable
quantities can be found, except for some quantities pointed to by global
low segment static locations at offsets YCSZAC, YOBJAD, YTXZTV, YSYSIN, YSYSOU, etc.
.s;X16
.br;
X16 is called XIAC,XFP, or XLOW. XIAC is used to point to some block
other than XCB when a variable is accessed. The contents of XIAC are regarded
by the compiler as lost whenever a change of control can occur
as when calling a procedure, at an explicit or implicit label, when
intermediate results may have to be saved, and when calling mathematical
subroutines. Inside the RTS, the ac can be used as XLOW to point to
the static area in the low segment. The UNIVERSAL file
SIMRPA contains a macro LOWADR which loads a specified
register with this address, and assigns the name XLOW to the register.
If no parameter is given, the default XLOW=XIAC is used.
The macro SETLOW(x) assigns the name XLOW to x (default XIAC if no
parameter is given), but does not load the register. Should be used when
it is known which register contains the address.
The name XFP is used on X16 in connection with FORTRAN subroutines, as
shown in the previous section. The standard calling sequence in
FORTRAN-10 is:
.nofill
MOVEI XFP,arg
PUSHJ XPDP,routine
where the argument list has the form:
XWD -number of arguments,0
arg: Z type code,address of first argument
Z type code,address of second argument
etc.
.fill
.justify
The result is returned in X0 and X1 if applicable.
.s;X17
.br;X17 is only referred to as XPDP and used for all PUSHJ-POPJ calls.
It refers to the run-time stack whose bottom is at YOBJRT, so named
because the return address to SIMULA code is placed there when
calling a RTS procedure.
.p 0,4,6;III.1.4.3##Coding conventions
.s;III.1.4.3.1##Preserving registers within the RTS
.s;Inside the RTS, all work registers can be used
provided any possibly relevant registers are preserved.
When some storage allocation routine is called, the calling RTS routine is
responsible for preserving the integrity of object pointers which may
exist in registers temporarily.
These registers must be saved in locations
which are known to the garbage collector as possible object references.
The array YOBJAD in the low segment is provided for that purpose.
Any unused cells in YOBJAD must be zero when not in use.
Other global cells known to the garbage collector are used for relocatable
address esin the SIMULATION module (SU) and in SIMDDT.
For a temporary text variable, YTXZTV should be used in the same
fashion. Registers which contain non-relocatable values
will be saved and restored by the garbage collector.
.p 0,2,10;III.1.4.3.2##Addressing the low segment
.s;The low segment is addressed through the right half of .JBOPS. As explained
above, the macros LOWADR and SETLOW are used to establish the base
register, which is always called XLOW (renamed by the macros).
The storage allocating routines in particular always use the
standard XLOW=XIAC to prevent confusion when calling them from other
RTS routines.
The use of global variables is avoided when the
stack can be used equally well.
.p 0,2,10;III.1.4.3.3##Saving intermediate results
.s;The RTS procedures which are of a function nature, i e can be used inside
an expression, warrant special consideration because of the requirement
to save intermediate results over the execution of the procedure.
If the procedure involves storage allocation, an acs (ZAC) object must be
created for the intermediate results. This is done in either of two ways.
.s;1) The entry point .CSSA is called from compiled code in the following
way:
.nofill
PUSHJ XPDP,CSSA ;via transfer vector
XWD N,ADMAP
.fill
Only the accumulators below the top ac (as given by YTAC) are saved.
This method cannot be used when the top accumulators are important.
_.CSSA creates a ZAC object, saves the intermediate results and
sets YCSZAC(XLOW) to point to the acs object.
YCSZAC is then used by a run time setup routine (.CSSN, .CPNE) to
set ZDRZAC of the display vector. YCSZAC is cleared after use.
.s;2) The "XWD N,ADMAP" word is passed as an inline parameter to some
RTS routines which may allocate storage directly or indirectly (e g
through the actions of a thunk). It is the responsibility of
such a routine to preserve any quantities which will not be saved
in the acs object, in the special locations reserved for object addresses or
text variables, then to call .CSSA. (secondary entry point to .CSSA) with
a copy of the inline parameter in XSAC, provided it is non-zero. A zero
inline parameter signifies that no intermediate results exist.
.br;When returning from an object generator, procedure, thunk evaluation
or some run time function such as .TXCY, YCSZAC is recovered and
_.CSRA is entered if YCSZAC is non-zero. The result of the
function is in XRAC _& XRAC+1 and is placed in the proper position on the
stack by .CSRA before restoring the intermediate quantities.
The result is thus always placed in the top accumulators.
.ts 1,9,17,25,33,41,49,57,65
.st 741111####780302######6###########################Lars Enderin
.pg;.lm 9;.i -8;III.1.5#RTS storage allocation
.br;----------------------
.p -8,2,10;II.1.5.1#Types of records allocated by the RTS
.s;Several types of dynamic blocks are allocated by the RTS in response
to the expressed needs of a SIMULA program. The different block types
are:
.nf;.s;
Record Function Allocated by
name
ZBI block instance SAAB (via CSSB)
(unreduced subblock)
ZBP procedure block SADB (via CSSN or CSSW
or CSSW0)
ZCL class object SADB (via CPNE)
ZPB prefixed block SADB (via CPSP)
ZTE text record SAAR (via TXBL, TXCY, IOIT)
ZTT temporary text variable SAAR (via TXDA)
ZAR array record SAAR (via CSNA or CSCA)
ZAC accumulator stack rec. CSSA
ZER eventnotice record SAAR (via SANE)
ZDR display record SADB (via CSSN, CSSW, CSSW0,
CPNE, CPSP)
ZYS "system" record SAAR
ZXB extended lookup block SAAR (via IOCF)
.f;.p 0,1,1;The block layouts are documented in TD I.5 and LH2 appendix
B.
.p -8,2,10;III.1.5.2#Some notes on the various block types
.s;The ZDNTYP field in the first word of all these blocks shows the
block type. The numeric values of block type codes are determined
by an expansion of the TYPZDN macro in SIMMCR.MAC (III.5).
The symbolic names are formed from the record names by prefixing
the letter Q, e.g_. QZCL is the assembly constant for the type code
of a class object record.
.p -8,2,10;III.1.5.2.1#Subblocks (ZBI)
.s;ZBI (unreduced subblock instance) is the simplest type of independent
program block. The block types ZBP, ZCL and ZPB all contain a ZBI
part. A ZBI block may contain several reduced subblocks corresponding
to Algol blocks at inner block levels.
.br;The ZBIBNM field in the block instance indicates the innermost
currently active reduced subblock.
The value in ZBIBNM is used by the garbage collector to find the
correct block map(s) (see I.5) and by SIMDDT to find the correct symbol
table segment.
.br;Unreduced subblocks are only allocated in the following cases:
.br;* A subblock immediately enclosed by a class instance. This saves
space for detached class objects by allowing the subblocks to be garbage
collected.
.br;* The outermost block in a connection statement.
.br;In all other cases, subblocks are reduced into the nearest enclosing
instance of a procedure, prefixed block or unreduced subblock.
Reduction of subblocks saves storage allocations, reduces the size
of display vectors (since several subblocks can be addressed from
the same base address) an speeds up execution. In return, some run
time procedures become more complicated, e.g_. CSGO.
See III.1.6.
.p -8,2,8;III.1.5.2.2#Procedure blocks (ZBP)
.s;Procedure subblocks (ZBP) are set up by calls to CSSN, CSSW or
CSSW0. They have an attached display vector (ZDR). SADB allocates
both ZDR and ZBP at the same time as one contiguous storage block.
.br;SADB also fills in those fields which are common to the block types
handled (ZDR + ZBP, ZCL or ZPB), e.g_. the reactivation point
(ZDRZBI,ZDRARE).
.p -8,2,10;III.1.5.2.3#Class objects (ZCL), prefixed blocks (ZPB)
.s;Class objects (ZCL) and prefixed blocks (ZPB) are allocated in
the same fashion as procedure blocks by calls to CPNE or CPSP, which
call SADB.
.p -8,2,5;III.1.5.2.4#Text records (ZTE)
.s;Text records (ZTE) are allocated by SAAR. They contain
the ASCII representation of the text contents.
.p -8,2,8;III.1.5.2.5#Temporary text variables (ZTT)
.s;ZTT records are created in connection with parameter handling.
A ZTT record contains a text variable (ZTV instance) which is used
as stand in for a text expression in certain cases. See parameter
handling.
.p -8,2,10;III.1.5.2.6#Array records (ZAR)
.s;ZAR records (arrays) are created by array declarations or when
passing array parameters by value. CSNA is used to create the first
array in an array segment. CSCA is used to create the other arrays
in the segment by copying and also when passing an array parameter
by value.
.p 0,0,8;Before the array record is allocated in CSNA, its size must
be computed from the subscript bounds. The subscript bounds are in
ac's (possibly also in pseudo ac's) and are saved in a ZAC record
during evaluation. Temporarily, the array thus requires its own space
plus that needed by the ZAC record (4 + 2n words, where n is number
of subscripts).
.p -8,2,10;III.1.5.2.7#Accumulator stacks (ZAC)
.s;ZAC records (accumulator stacks) are created in situations where
intermediate results are in ac's and possibly in pseudo ac's, and
the garbage collector may be called because of storage allocation.
.br;Since space is always reserved at the top of the storage pool
for a ZAC record of maximal size, CSSA can create the record and copy
the ac values into it without first checking the storage limits. If
sufficient space for another ZAC is not left then, the garbage collector
is called to provide more space.
Relocation information for the saved values is provided in the accumulator
map pointed to by the ZACZAM field of the ZAC. The ac map is a bit
vector where a one in bit position n-1 signifies that saved result
n (XWACn) contains an object address in its right half, e.g_. a REF
value, the first word of a TEXT variable, an array address, a procedure
dynamic address, etc.
.br;If pseudo ac's are used, relocation bits are found starting with
bit 18 of the map word.
.br;The address of the ZAC object is placed by CSSA in YCSZAC(XLOW)
whence it may be copied to another location (e.g_. ZDRZAC of a display
vector). YCSZAC(XLOW) will be cleared when it is no longer needed.
ZAC records may be created when the following RTS routines are called:
.br;CSSA (before a call to CSSN or CPNE), CSSW, PHFA, PHFV, PHFM,
PHFT, TXBL, TXCY, TXDA, IOIT, CSNA.
.br;The other procedures call CSSA at its entry point _.CSSA_. .
.br;Note that the current "top" ac's are NOT saved in the ZAC object.
Instead, they are saved by the invoked RTS routine in global locations
provided for that purpose, if pointers are involved,
or on the RTS stack for quantities which are invariant under
garbage collection.
.p -8,2,10;III.1.5.2.8#Display records (ZDR)
.s;ZDR (display) records exist for procedures, non-terminated class
objects and for prefixed blocks (including the main program block).
For classes with local classes as attributes, the display record is
kept even when terminated because it may be needed to establish the
environment of a procedure attribute of a local class.
.p -8,2,6;III.1.5.2.8.1#Static display
.s;The ZDR record contains the traditional static display known from
Algol implementations. Instead of one central display vector, we have
here one display for each active block of a type described above.
This means that several pieces of information are duplicated in the
system. Because most subblocks are reduced, however, quite few display
levels are usually needed, and displays for terminated class object
are usually deallocated. Since the display is adjacent to the block
and addressed via the same ac (usually XCB, the current block pointer),
one instruction usually suffices to obtain the address of any block
in the static environment.
.p -8,2,6;III.1.5.2.8.2#Dynamic link
.s;In addition to the display, ZDR also contains the dynamic link
or reactivation point (ZDRZBI,#ZDRARE). For a procedure or attached
class instance, the reactivation point is (address of calling block
instance, return address). For a detached class instance, ZDRZBI normally
points to its own block, and ZDRARE points to the reactivation point
within its code (program point after call on DETACH or RESUME or after a
scheduling
statement in SIMULATION). See III.1.6.3.5 Quasi-parallel Sequencing.
.p -8,2,6;III.1.5.2.8.3#FOR loop returns, thunk save areas, ZAC pointer
.s;Return addresses for FOR loops are also contained in the ZDR record,
as well as thunk save areas (see parameter handling), and the address
of a ZAC record if one was created just before the ZDR record was
created. The existence of a ZAC record for the attached block is flagged
by a bit in that block instance to prevent execution of unnecessary
instructions if a ZAC record does not exist. This is relevant because
the ZDRZAC field is accessed from the front end of the ZDR, and the
front end address has to be calculated from the block address and
prototype information, whereas the other information in the ZDR record
is accessed conveniently via negative offsets from the block address.
.p -8,2,6;III.1.5.2.9#Eventnotice record (ZER)
.s;ZER (eventnotice) records are used in SIMULATION programs. Each
SIMULATION block has one or more ZER records allocated, chained via
the ZSUZER field in the SIMULATION block and the ZERZER field in the
ZER record. A ZER record contains a number of event notices (ZEV).
.p -8,2,6;III.1.5.2.10#Eventnotices (ZEV)
.s;Each ZER record contains several ZEV records (eventnotices), of
which some are members of the sequencing set (SQS).
The free ZEV records form a chain via their ZEVZEV fields and the
first free record is found via the ZERZEV field in the ZER object.
For garbage collection purposes, the ZEVZER field of each ZEV gives
the address of the ZER record of which the ZEV is a part. The ZEV
also contains links to other ZEV instances in the SQS, a pointer to
the associated process (ZEVZPS) and the scheduled simulated time for
the next active phase of that process. The process has a pointer to
its associated ZEV record (ZPSZEV).
.p -8,2,8;III.1.5.2.11#ZYS records
.s;The ZYS record type is used for various blocks "behind the scenes",
i.e_. not directly related to any SIMULA concept. A ZYS block may
only contain information invariant under garbage collection, so that
it can be moved freely without relocation of internal information.
ZYS blocks are used e.g_. for sub-file directories (SFD), and SIMDDT
is marked as a ZYS record for garbage collection purposes.
.p -8,2,6;III.1.5.2.12#Extended lookup/enter block (ZXB)
.s;ZXB records are used for extended lookup/enter blocks in the SIMULA
I/O subsystem. A ZXB record may contain a pointer to a sub-file directory
and is thus not invariant under garbage collection.
.pg;.p -8,0,10;III.1.5.3#Garbage collector (GC)
.s;The garbage collector (GC) can be called for three reasons.
.lm 12;.p -3,1,3;1.#The storage pool is exhausted, a new core request cannot be
honored within the allocated core (top location given by YSALIM).
.p -3,1,3;2.#New buffers are to be allocated. The whole pool must be moved
upwards, leaving free space at the bottom.
.p -3,1,3;3.#SIMDDT calls the garbage collector when executing the VARIABLES
command in order to remove garbage data before dumping the pool.
.lm 9;.p 0,1,5;In the first case ac 0 (X0) contains the number of
words required above the limit given by YSALIM(XLOW). In the second
case, YSAREL(XLOW) contains the number of words needed at the bottom
of the pool, and in the third case X0 and YSAREL are both zero, thus
no more core is needed.
.p 0,1,5;The action taken when GC is called (entry .SAGC in SA.MAC)
depends on the allocation strategy chosen when assembling the RTS.
If the assembly constant QSASTE is non-zero, the pool is allocated
in steps (i.e_. the pool is expanded by one or more core requests
between garbage collections rather than collecting garbage each time).
If a new core chunk (step) of a previously calculated size can be allocated
without
exceeding the garbage collection limit YSAL(XLOW) calculated in an
earlier execution stage, the step is taken, otherwise a complete garbage
collection is performed. If X0 was zero on entry, however, a complete
garbage collection is always performed (cases 2 and 3 above).
.p 0,1,10;If QSASTE is zero, the whole pool given by the value in
YSAL(XLOW) is allocated initially and after each garbage collection,
and a call to SAGC will always result in a complete garbage collection
(i.e_. the limits YSALIM and YSAL coincide). This will probably be
the best strategy for KA10 installations where CORE requests are fairly
expensive. QSASTE is defined in SIMMAC.MAC and set to 1 as default.
A change of the QSASTE value only affects the SA.MAC module which
must be reassembled after assembling SIMMAC.MAC.
.p 0,1,10;Another algorithm has been added to take care of big programs
which may run out of real core memory and start to use virtual memory.
After calculating the optimal YSAL value, check to see if the job
would go virtual. In that case, modify the estimated limit to a value
which avoids excessive paging overhead. This algorithm is used only
when virtual memory is available and it is currently not well tested,
but it has reduced execution time significantly for a couple of programs.
.p 0,2,5;Garbage collector phases
.s;The garbage collector works in four phases.
Figure III.1.5.3 shows the interaction of GC components (subroutines
and coroutines).
.pg;Figure III.1.5.3 Subroutine and coroutine linkage in the
garbage collector.
.nf;.s;.lt 43;
+------+ +------+ +------+ +------+
!.SAGC1!--->!PHASE1!--->!SAGCGP!<-->!SAGCNP!
+------+ +------+ +------+ +---:--+
! ^ V
! ! +------+
! +-----!SAGCCH!
+------+ ! +------+
!SAGCDR! V
+------+<-->+------+ +------+
!SAGCSP!<-->!SAGCNP!
+------+<-->+------+ +--:---+
!SAGCFP! ^ ^ V
+------+ ! ! +------+
V +-----!SAGCCH!
+------+ +------+
+ ------------------!SACGN1!
! +------+
V
+------+ +------+ +------+ +------+
!PHASE2!--->!PHASE3!--->!SAGCGP!<-->!SAGCNP!
+------+ +------+ +------+ +---:--+
! ^ V
! ! +------+
! +-----!SAGCUP!
+------+ ! +------+
!SAGCDR! V
+------+<-->+------+ +------+
!SAGCSP!<-->!SAGCNP!
+------+<-->+------+ +--:---+
!SAGCFP! ^ ^ V
+------+ ! ! +------+
V +-----!SAGCUP!
+------+ +------+
+ ------------------!SACGN3!
V +------+
+------+ +------+
!PHASE4!<->!.SANP !
+---:--+ +------+
V
+------+
!RETURN!
+------+
.el;.f;.pg;.i-8;III.1.5.3.1#Phase 1
.s;The first action is to find all referenceable records in the pool.
The right half of the first word of all records in the pool is reserved
as GC working space (ZDNLNK field). During phase 1 this field is used
to chain all referenced records. The chain starts with the outermost
block instance, which can be found provided XCB points to a record
with an attached display record. The outermost block instance is part
of generated code for the main program and is thus not placed in the
dynamic storage pool. Usually, though, the outermost block contains
pointers into the dynamic pool which may have to be updated during
garbage collection. If no display record is attached to the XCB record,
the chain starts with the XCB record itself. All global pointers in
the static area (defined in SIMRPA, macro STATIC) which may contain
pointers into the pool are checked, and each record found via these
pointers is attached to the chain. Since a dynamic record in the pool
may contain other pointers, each record in the chain must be searched
for pointers. Phase 1 is completed when the last record of the chain
is checked without finding any new pointer.
.p -8,2,6;III.1.5.3.2#Phase 2
.s;At this stage there are two kinds of records in the pool:
.br;* Referenced records with a link address in ZDNLNK.
.br;* Unreferenced records with ZDNLNK = 0.
.br;New record addresses can now be computed for all referenced records,
assuming that the records should be moved towards the bottom of the
pool in the same order by increasing addresses.
If YSAREL(XLOW) is non-zero, its value is added to the old bottom
address. Then the whole pool is scanned, placing the new address of
each referenced record in its ZDNLNK field. Unreferenced neighbours
are lumped together to look like a single unreferenced record in the
following phases.
.p 0,0,5;When all referenced records have been assigned new addresses,
the new value of the first free location (the new YSATOP value) can
be determined. This plus the amount of core requested when
calling GC gives the minimal amount of core required for continued
execution of the SIMULA program.
If not enough core was reclaimed by the GC, a CORE UUO is executed
for the required core size. If that fails, execution terminates with
an error message.
.p -8,2,6;III.1.5.3.3#Phase 3
.s;All pointers that were checked during phase 1 must again be checked.
This time any pointer into the pool will be updated to point at the
updated location of the record (given by its ZDNLNK field).
The same routines are used as in phase 1, but the records are not
treated in the same order. In phase 3, referenced records are treated
in the order in which they appear in the pool, starting at the bottom.
Instead of calling SAGCCH for each record, SAGCUP is called to perform
the updating of addresses. See figure III.1.5.3.
.br;Some dynamic records contain internal pointers. These are relocated
as soon as the next record in the pool is found by the SAGCN3 routine.
Eventnotice pointers are also treated here. They are found in SIMULATION
and PROCESS block instances and in eventnotice records.
.br;At this stage, however, ZEVZER and ZERZER pointers cannot be updated
since they are used to define the relocation offset for eventnotice
pointers not yet found.
.p -8,2,6;III.1.5.3.4#Phase 4
.s;Now the ZERZER chain and all ZEVZER pointers can be updated. Each
referenced record in the pool should now be moved to its new location
given in the ZDNLNK field. A complication occurs if the bottom of
the pool should be moved upwards. For some records moved towards
a higher address, the old and new area may overlap, prohibiting the
use of a BLT instruction. In that case, the record is moved word by
word starting at the high end, and the records must be moved in reverse
order. This is simplified by first making a reverse chain of records via
the ZDNLNK fields.
After all records have been moved to new locations, the reclaimed
core is cleared to zeros.
.p -8,2,8;III.1.5.3.5#Determine and allocate new storage pool area
.s;After phase 4, a new garbage collection limit and a new allocation
step size if applicable, are calculated. The algorithms used in the
calculations are given below.
.br;Finally, a CORE request is made if necessary. The maximal available
amount is used if the request is excessive.
.p -8,2,6;III.1.5.3.6#Core size algorithm
.s;The core size algorithm is designed to minimize running costs for
the SIMULA program assuming an accounting formula of the following
form:
.s;####Cost = G * (L + A) * time
.s;where L is the storage pool size, G is a constant, and A is explained
below.
.br;It is further assumed that the time required for a garbage collection
is proportional to the amount of active memory. The proportionality
factor B is approximated with a divided difference over the last garbage
collection. It is also assumed that the allocation rate R is approximated
by a divided difference from the end of the last garbage collection
to the start of the current garbage collection. In the KI10 production
system core is released after a garbage collection and reclaimed when
needed in constant steps, each step an integral number of 512-word
pages (see next section).
.p 0,1,3;With the current accounting algorithm, A can be approximated
by
.p 0,1,6;.nf;
# 5450 Q-La
# A = --------- + 5 + ---- pages,
# (La+Q+10) 2
.f;.p 0,1,6;where Q is the size of high segment and code and data outside
the storage pool, and La is the average size of the storage pool or
(F+L)/2, where F is the size of active memory and L is the previous storage
pool limit. The following figure shows how A was computed:
.nf;.p 0,1,30;A = dA/kA.
.s;
Cost/second
.s 2;
_^
!
! .
! + .
! + .
! +.:
! : + : Slope of tangent =
! Accounting + : kA = G
! algorithm +: : :
! ! : + . : : :
! v + . : : :
! + :--- : : :
+ ! . : ! : : :
+ ! . : ! : : :
.! : ! : : :
. ! : dA : : :
. ! : ! : : :
. ! : ! : : :
. ! : ! : : :
---------------+-------+-------+---+------+-----> Allocation
:.......:.......: : :
Q : F : :
:...........: :
: La :
:..................:
L
.f;.p 0,1,5;The optimal storage pool limit minimises the sum of the
cost to execute between garbage collections and the cost to execute
the garbage collections. Assuming stationary program dynamics
(constant allocation rate R, constant active memory size F, constant
garbage collection time per word B and infinite execution time) the
new optimal storage pool limit is:
.tp 4;.s;.lt 3
A
L = F * (1 + sqrt[2*B*R*(1 + -)]) (1)
F
.el;.s;The expected time between garbage collections is (L-F)/R,
the average store size is (L+F)/2, which yields the cost between garbage
collections:
.tp 4;.s;.lt 3
L-F L+F
--- * (--- + A) (2)
R 2
.el;.S;The expected time for garbage collection is F*B, and the cost is
.nf;.s;
####F*B*(L+A) (3)
.f;.s;The cost rate per effective cpu second is thus
((2)+(3))/((L-F)/R), and minimisation with respect to L gives (1).
.p -8,2,10;III.1.5.3.7#Step size algorithm
.s;Assume that the accounting algorithm can be written
.s;####COST = const * [A+W*(M+U)*M]*time
.s;where M is accounted memory.
.p 0,1,5;Assuming approximately linear memory size dependency for
this execution phase we have
.s;####COST = const * [A+(B*M+C)]*time
.p 0,1,6;Assuming average time for a CORE UUO is K, allocation rate
is R and that core is expanded in the interval (C0, C1), we can derive
an approximation for the optimal step size:
.tp 4;.s;.lt 3
A+C
S = sqrt(K*R*[2 * --- + C0 + C1])
B
.el;.nf;.p 0,1,6;Define m = W*(M+U)*M
We then have
.tp 4;.s;.lt 3
dm
-- = 2*W*M+W*U
dM
.el;.s;and the linearisation yields
.s;####m = W*(2*Ma+U)*M-W*Ma_^2
.s;where Ma is the average memory size.
.s;Thus B = W*(2*Ma+U)
and C = -W*Ma_^2
.s;Define X = 2*Ma = C0+C1
.p 0,1,3;Thus B = W*(X+U)
C = -0.25*W*X_^2
.s;and the optimal step:
.p 0,1,5;.lt 3
2*A-0.5*W*X^2
S = sqrt(K*R*[------------- + X])
W*(X+U)
.el;.s;The bracketed expression can be expressed as:
.p 0,1,7;.lt 5
2*A X^2 2*A X^2
--- - --- --- - --- + X^2 + X*U
W 2 W 2
--------- + X = --------------------- =
X+U X+U
.el;.p 0,1,7;.lt 5
X^2 2*A 2*A U^2
--- + X*U + --- --- - ---
2 W W 2 X+U
= --------------- = --------- + --- =
X+U X+U 2
.el;.p 0,1,7;.lt 5
4*A
--- - U^2
1 W
= - [--------- + X+U]
2 X+U
.el;.p 0,1,6;.lt 5
The present values at our installation are
U = 20 [pages]
A = 1.1 [1/sec]
W = 0.0001 [1/(sec*pages^2)]
.el;.p 0,1,10;Thus
.p 0,0,5;.lt 3
K*R 44000 - 400
S = sqrt(--- * [----------- + X+20]) [pages]
2 X+20
.el;.p 0,1,8;Expressed in words and milliseconds and assuming K = 4 ms
we will get
.p 0,0,5;.lt 3
1.143*10^10
S = sqrt(2*R*[----------- + X+10240]) [words]
X+10240
.el;.p 0,1,7;.lt 6
where
R = allocation rate in words/ms
4*A
1.143*10^10 = [--- - U^2]*512^2 [words*words]
W
10240 = U * 512 [words]
.el;.f;
.p -8,2,15;III.1.5.3.8#Algorithm for virtual memory
.s;The algorithms described above are not suitable when the
job goes virtual, i.e_. when the available real core memory
is not sufficient for the job to run. If the user has virtual
memory privileges, the job will then start paging. In this
case, the assumptions on which the above algorithms are based
will not be valid. Essentially, use of virtual memory
should be avoided.
An alternative allocation algorithm has been constructed for the case
where virtual memory is used. The assumptions on which the algorithm
is based may not be quite valid, but significant cost improvements
have been obtained for programs using much virtual core.
.p 0,1,5;- At start of GC, save number of niw faults in YSANWA
and number of niw faults since last GC in YSANWB.
GETTAB [%VMSPF] used. Cumulative count of niw faults
updated in left half of YSANWC.
This should be job-related data, but only global system counts are
available in this way, which means that other jobs executing at the
same time may add to the counts, causing more restrictive use of memory
than necessary.
.p 0,1,5;- Modification of core limit if CORE fails in SANP1:
If first CORE fails, check if job is going
virtual, and if it does, subtract space (1K) for PFH.
GETTAB [-1,,.GTCVL] used.
.p 0,0,3;- Determination of new allocation limit for
a job that runs in virtual core:
.br;In SANP, if the page fault handler is in core, determine number of niw
faults between last two GC:s (from YSANWB) and change
YSAL by 2K*SIGN(tgc-tswap*(nb+2*ng). Here
.br;nb niw faults before last GC;
.br;ng niw faults during last GC;
.br;tswap time (ms cpu) accounted for a page swap
(currently = 20.);
.br;tgc time spent in this GC.
.br;However, if YSAL becomes less than minimal quantity
needed, the latter is passed to SANP1.
The pool is not allowed to exceed the low segment limit (128K).
.p 0,0,3;- If job is not virtual, but the new estimated
optimal YSAL value will make it virtual, make the new
YSAL value so that the job just goes virtual:
.br;(YSAL=min(physlim+qpolmi,virtlim)-YSAHSZ-sizeofPFH).
.br;GETTAB[-1,,.GTCVL] is used to find out virtlim.
.pg;.c;[This page intentionally left blank]
.pg;.c;[This page intentionally left blank]
.pg;.c;[This page intentionally left blank]
.pg;III.1.6##Execution control
.br;#########-----------------
.p 0,2,10;The execution of a SIMULA program is controlled on several levels -
by the user through monitor commands, by the RTS, and by compiled code.
.s 2;1) Monitor commands - START (CSTART), REENTER, CONTINUE (CCONTINUE), _^C.
.s;2) Overall program control by the RTS:
.br;High segments are swapped by OCSW.
.br;A program is set up and initialized by OCSP, OCIN, OCEI.
.br;Errors are handled by OCUU and SIMDDT, traps by OCTR and OCUU.
.br;SIMDDT interface is handled by OCUU, .OCRE0, .OCRE, .OCLD, .OCRD.
.br;OCEP terminates program execution.
High segments are swapped by OCSW.
.s;3) Procedure calls are handled by CSSN, CSSW, CSEN, CSEP, CSES.
Class objects are created and controlled by CPNE, CSEN, CPCD, CPCI, CPE0.
Subblocks are created by CSSB and terminated by CSEU.
Reduced subblocks are initialized by CSER.
Prefixed blocks are handled by CPSP, CSEN, CPPD, CPE0, CPCI.
Coroutine sequencing and quasi-parallel systems are handled by CPDT (detach),
CPCA (call), CPRS (resume).
SWITCHes and GOTO statements are handled by CSSC, CSES, CSGO.
Evaluation of a name parameter with a thunk may cause control to go anywhere.
See III.5 (parameter handling).
SIMULATION is handled by special statements
and procedures - see III.1.10.
.s;4) Conditional expressions and statements, WHILE loops, FOR
loops and simple GOTO statements are handled entirely by compiled code.
.p 0,3,6;III.1.6.1##Monitor commands for SIMULA program control
.s;III.1.6.1.1##START (CSTART)
.skip !
A START or CSTART command causes a SIMULA program in core to start execution
at the entry point given by the right half of .JBSA.
LINK-10 sets the start address to .MAIN if a LOAD or EXECUTE command was issued.
If COMPIL recognizes the program as a SIMULA program or the /SIMULA switch
is given, a DEBUG command causes LINK-10 to set the start address to .OCRE0,
otherwise DDT will usually be loaded and the start address will be .MAIN.
To understand what happens at the start of a SIMULA program it is necessary
to know the general layout of a SIMULA main program, as shown below:
.s;The first 4 or more words are copied from the
source file LOOKUP information.
The interesting information is filename.ext, date and time
of file creation and last access, and the access path:
.lm 16
[proj,prog] (zero if no path given) or [proj,prog,sfd1,...]
if SFD files are part of the directory path.
See Monitor Calls manual for lookup block and SFD information.
.s;.nofill
[source file lookup info]
[text constants and any runswitches block]
[Display (ZDR) for main program block]
.i -8;mainb:##[Block instance (ZPB) for main program]
[constants]
.i -8;_.MAIN:##JRST 1,.+1 ;Normal entry point
TDZA X1,X1 ;No SIMDDT address
JRST 1,.+1 ;Enter here from .OCRE0
MOVEI XCB,mainb
JSP X16,.OCSP
JFCL runswitches address ;may be zero
[compiled SIMULA statements]
PUSHJ XPDP,OCEP
[prototypes, line number tables, symbol tables]
.s;.lm 8
.fill
If the program is started at .MAIN, X1 is set to zero to indicate
that SIMDDT is not in core. The program will then be started without SIMDDT,
unless this is a restart after program exit and SIMDDT was loaded
at the start of some previous execution. In that case, the address of
SIMDDT has been placed in YOCDDT, a global location in OCSP. See REENTER
command (III.1.6.1.2).
OCSP loads the initial high segment after allocating core for the static
data area and saving the ac's. Control is then transferred to OCIN for
the bulk of the initialization - setting up the RTS stack,
allocating standard files SYSIN and SYSOUT and buffers, reading
any specification file, etc.
OCIN finishes by transferring control to OCEI in the OCEP module. If the
two high segments version of SIMRTS is used, invocation of OCEI from OCIN
leads to a high segment swap via the transfer vector and .OCSW.
When the program starts, a RESET UUO stops any active I/O and the buffer
space is released by moving .JBFF back to the value in .JBSA(LH).
Any active SIMDDT breakpoints are removed by OCSPDR.
.s;CSTART is identical to START except that the terminal remains in monitor mode.
The program will probably wait for input or output on the terminal unless
continued via _^C-CONTINUE.
.p 0,5,6;III.1.6.1.2##REENTER command
.s;The REENTER command has different effects depending on the current
state of execution in the SIMULA program. The current address in .JBREN
governs the actions.
.p 0,2,10;III.1.6.1.2.1##REENTER before program start
.s;If issued directly after a LOAD or GET command, REENTER causes SIMDDT to be
loaded and entered before executing the statements of the SIMULA program.
REENTER starts the program at the entry point .OCRE0 in the OCSP
module.
If COMPIL does not recognize the program as a SIMULA program, a DEBUG command
will cause DDT to be loaded instead of SIMDDT. A _^C followed by REENTER
will have the same effect as REENTER after LOAD in this case.
_.OCRE0 reserves space for SIMDDT if not already loaded, modifies .JBSA(LH) to
keep this space over RESET, and transfers control to the compiled program at
_.MAIN+2 with X1 containing the address of the SIMDDT area. The first word of
that area is set to zero if SIMDDT was not loaded previously.
.br;In OCSP, the assigned SIMDDT address is saved in YOCDDT for later use (restart).
When OCEI eventually gets control from OCIN, SIMDDT is loaded if necessary and
entered at the entry point DSINI. Breakpoints may then be set before starting
execution of compiled statements via a PROCEED command.
.p 0,2,7;III.1.6.1.2.2##REENTER during RTS initialization
.s;If the program is interrupted while _.OCSP, .OCRE0 or OCIN is active,
a REENTER command has the same effect as a CONTINUE command except for typing
a message on the TTY. The entry point OCRE1 is used in that program phase.
.p 0,2,10;III.1.6.1.2.3##REENTER during execution of SIMULA statements
.s;During normal SIMULA program execution, the reentry address is .OCRE.
If the program is stopped by one or two _^C characters in this phase,
a REENTER command causes .OCRE to be entered.
Depending on the value of the control variable YDSCSW(XLOW), SIMDDT may
be entered
.br;a) directly,
.br;b) after returning to some well-defined point in the program,
.br;c) or the request to enter SIMDDT must be denied because the data structures are
being changed so that they are possibly inconsistent at the time of
interrupt.
.br;In case (c), a new interrupt may be attempted after running
the program for some time. If SIMDDT was not loaded with the program,
it is loaded from the file SIMDDn.ABS
(n = SIMULA version number).
Once within SIMDDT, breakpoints may be set, variables and the operating chain
may be displayed, but some commands can not be
performed, chiefly because garbage collection may be involved.
.p 0,2,10;III.1.6.1.2.3.1##Basic requirements for allowing SIMDDT to be called
.s;The following basic requirements must be met when SIMDDT is executed in this
mode:
.s;a) No garbage collection allowed
.s;Since there is no way to determine which registers and temporary cells contain
pointers into the SIMULA storage pool, garbage collection cannot be allowed
directly or as a side-effect, i e no storage may be claimed from the pool
without making sure that no garbage collection will result.
.s;b) The data structure must be consistent
.s;The data structures used by SIMDDT must be consistent in order not to get wild
results or cause program interrupts. In particular, the current block pointer,
XCB, must be consistent with the program address used
by SIMDDT to pinpoint the program state (program point and current data).
XCB may not point to a block which is partly correct, e g does not have a proper
display vector.
If the interrupt was not in compiled SIMULA code, the bottom stack level must
point to a program point consistent with XCB.
.s;c) SIMDDT may not call non-reentrant routines which change
global data in use at the time of interrupt
.s;This requirement is met by not permitting SIMDDT to be called while such
a routine is active.
.p 0,2,10;III.1.6.1.2.3.2##Controlling the effect of REENTER
.s;In order to conform to the basic requirements,
special control features have been
implemented for determining when and if SIMDDT can be called.
For the purposes of this description, a SIMULA program with its attendant run
time system (RTS), possibly linked with some FORTRAN
or MACRO-10 subroutines, is in one of three states:
.s;State A (allow).
.br;In this state, SIMDDT may be invoked directly.
The data in the pool is consistent with the program point given by XCB and the
interrupt address (usually .JBOPC).
A SIMULA program is in this state when it is at code level, i e outside
any RTS routines in code compiled by SIMULA. State A also applies
to some (sections of) RTS routines, e g mathematical subroutines.
The bottom of the stack (YOBJRT) then points into the SIMULA code.
.s;State D (defer).
.br;In some RTS routines, invocation of SIMDDT must be deferred to some
future instant, normally the return to user code. If such a point can be
easily identified by reference to the push-down stack, the stack can
be manipulated to effect the invocation of SIMDDT at the proper time.
Routines of this type should always return to the point of call, or the
stack must be set up to allow for such treatment.
If a proper invocation point cannot be found easily, the state must be F.
.s;State F (forbid).
.br;In this state, invocation of SIMDDT is forbidden. The program is restarted
with a message that a new attempt to stop and REENTER should be made.
A SIMULA program is in this state when the data structure could be
inconsistent and/or the program point cannot be identified, etc.
State F is avoided if possible.
.p 0,2,10;III.1.6.1.2.3.3##Macros for controlling the effect of REENTER
.s;The state information is encoded in a global cell in the static low segment
area (based on the address in .JBOPS and addressed by XLOW, which is normally AC16).
This cell is called YDSCSW and is manipulated by the following macros:
.s;CALLOW
.br;This macro sets the state to A by clearing YDSCSW(XLOW).
.s;CDEFER
.br;Adds 1 to YDSCSW(XLOW). State A goes to state D. Any other state is unchanged.
.s;CENABLE
.br;
Subtracts 1 from YDSCSW(XLOW). Will give state A if YDSCSW(XLOW) was 1,
otherwise the state is unchanged. Proper use of the macros will ensure
that YDSCSW(XLOW) is not 0 when CENABLE is executed.
CDEFER and CENABLE must be properly nested so that state A is reached when
an equal number of those macros have been executed, if the initial state was A,
and state F was not set.
.s;CFORBID
.br;Sets the left half of YDSCSW(XLOW) to -1. The state is thereby set to F
(YDSCSW negative).
.s;In order for these macros to work, the global cell YDSCSW must be
addressable via the register designated XLOW (set by the LOWADR or SETLOW
macro).
.p 0,2,10;III.1.6.1.2.3.4##Actions after REENTER
.s;If the program is interrupted and subsequently REENTER-ed, the RTS
actions depend on the state as follows:
.s;If the ordinary reentry point is not in effect, the state is irrelevant,
either because the RTS is not inintialized, or because .OCRE is active in
processing a REENTER command.
Execution merely continues with a message stating the reason.
.s;If the ordinary reentry point is in effect (.JBREN=.OCRE), the actions are:
.s;All registers are saved. .JBREN is changed to an alternative reentry point.
The PC value at interrupt (.JBOPC) is stacked as return address.
If .OCRE is entered from OCRET (deferred REENTER), .JBOPC is
the address of the return to SIMULA code, i e a faked PC value.
The state is determined and governs the actions.
.s;III.1.6.1.2.3.4.1##Calling SIMDDT directly after REENTER
.s;State A: Set a switch to disallow garbage collection.
If SIMDDT is not in core (loaded with program), bring it in if possible.
Start at the point DSINR of SIMDDT with XDSBAS set to the base of SIMDDT.
YDSCAD(XLOW) gives the program point of interrupt, if possible related to
compiled code.
When SIMDDT returns, the switch to disallow garbage collection is reset.
.s;_.JBREN is set to .OCRE, the registers are restored, and control returns
to the program via a "POPJ XPDP," instruction to where .JBOPC pointed.
.s;III.1.6.1.2.3.4.2##Preparing for deferred SIMDDT call
.s;State D: In this state, the interrupted instruction must be in some RTS
routine, and the stack must have the return
to compiled code at the bottom. Invocation of SIMDDT should be deferred
until the return point. This is achieved by fixing up the stack so that
instead of returning to compiled code, a special RTS routine will be entered.
In some cases, the return address at the bottom of the stack does not
point to the next instruction to be executed at return but to an inline
parameter. In that case, the inline parameter must be copied to a cell before
the start of the special reentry routine, and the cell at the stack bottom must
point to the copy.
The only inline parameters used in the present system at object code level are
of the form
.br;########XWD n,m
.br;where n is an integer whose absolute value is less than 512. Thus, the op code
part of the word is either 0 or 777 (octal), and the word cannot be an
executable instruction.
The code return address is saved in YDSCAD(XLOW), and the state is set to F.
The accumulators are restored and return is effected to where .JBOPC points.
When the special routine (OCRET) whose address has been put into the
bottom of the stack, is entered, YDSCAD(XLOW) is copied to .JBOPC,
state A is set, and .OCRE is entered.
The following actions are the same as for state A above.
.s;III.1.6.1.2.3.4.3##Call on SIMDDT not allowed
.s;State F: If REENTER is attempted in this state, execution is continued with a
message. A new attempt should be made after stopping the job again.
.p 0,2,6;III.1.6.1.2.4##REENTER while processing REENTER in .OCRE or SIMDDT
.s;The contents of .JBREN are OCRE2 in that case.
The current SIMDDT command is suppressed if possible.
If SIMDDT is executing,
_^C_^C followed by REENTER can thus be used e.g_. to suppress
the rest of a lengthy memory dump or source listing.
A message is typed and execution continued, or SIMDDT expects a
new command.
.p 0,2,6;III.1.6.1.2.5##REENTER after program exit
.s;This has the same effect as REENTER after LOAD or GET. When SIMDDT signals
that it is ready to accept commands, any old breakpoint information has been
reset.
.p 0,4,10;III.1.6.1.3##CONTINUE (CCONTINUE) command
.s;CONTINUE has the normal effect of continuing execution where it was stopped
by a _^C or by execution of some instruction such as EXIT 1, HALT, or
by some monitor error.
.s;CCONTINUE is identical to CONTINUE except for leaving the terminal in
monitor mode.
.s;When CONTINUE is given after normal program exit, it has the effect of
invoking SIMDDT as if a _^C-REENTER had been given just after the final END
of the main program, but with the block state of the main program block set
to 1 to allow examination of the global variables.
A MONRT. (EXIT 1,) UUO returns the job to monitor mode again.
.pg;III.1.6.2######Overall SIMULA program control by the RTS
.s;The modules involved in overall execution control are OCIN, OCIO,
OCEP and OCSP. OCIN and OCIO also contain some I/O code.
.p 0,2,10;III.1.6.2.1####Setting up and initialization of a SIMULA program
.s;A SIMULA program is started either directly by LINK-10 after
loading (EXECUTE or DEBUG command), or by a START or REENTER command
following a GET or LOAD command. The SIMULA program layout shown in
III.1.6.1.1 (START command) indicates the initial actions.
.p 0,2,10;III.1.6.2.1.1##Getting the high segment into core (OCSP)
.s;OCSP (Overall Control, Start Program) is the first component of
the RTS to be called from the SIMULA program. OCSP is loaded with the
main program, normally in the low segment. Since the low segment
cannot be shared, the functions of OCSP have been minimized, most
initialization functions being deferred to the the high segment. The
main function of OCSP is to load the initial high segment into core
via a GETSEG UUO, after allocation of a minimal amount of core for
the so called STATIC area of the low segment. If the high segment was
loaded with the program from REL files, and the main program was
loaded in the high segment, OCSP uses SETUWP to turn off write
protection, since the main program contains a data area. After
getting the high segment, OCSP transfers control to OCIN to perform
the rest of the initialization. The ac's are saved starting at offset
YACSAV in the STATIC area, after possibly getting the address of
SIMDDT from YOCDDT or from a REENTER call on .OCRE0. OCSP is called
by JSP instruction, saving the return address in an ac. In the
non-sys version, it is possible to look for the high segment on
several disk areas if the first GETSEG fails. See module listing for
further details.
.p 0,2,15;III.1.6.2.1.2##SIMULA program initialization (OCIN)
.s;OCIN starts with some preliminary initializations: Clears static
area, sets standard value for LINESPERPAGE and LOWTEN, records
time-of-day and runtime at program start for later use on program
exit, sets up the run-time stack (based on XPDP), finds any
RUNSWITCHES block, initializes trap and UUO handling. The rest of
OCIN is concerned with actions related to BASICIO, i.e_. SYSIN and
SYSOUT, specification of other files and allocation of i/o buffers.
This is described in III.1.7 (I/O facilities).
.p 0,2,15;III.1.6.2.1.3##End of initialization phase (OCEI)
.s;OCIN ends by transferring control to OCEI, which opens SYSOUT,
loads SIMDDT if space was allocated and transfers control to SIMDDT
or the first explicit SIMULA statement. If SIMDDT is first started,
it transfers control to the program after its own initializations.
OCEI was introduced to define a point where control passes from the
initial high segment to the other high segment if two segments are
used.
.p 0,2,15;III.1.6.2.2####Swapping of high segments (OCSW)
.s;The high segment parts of the SIMULA RTS can be classified into two
categories:
.br;a) Routines which are used frequently by an executing SIMULA
program, such as block creation and block entry, editing procedures,
etc.
.br;b) Routines which are used only at the start of execution or very
seldom during the execution of a normal SIMULA program.
.s;In order to reduce the memory space normally occupied by a SIMULA
program, the high segment routines (collectively named SIMRTS) are
grouped into two distinct high segments, SIMRn1 and SIMRn2 (n =
SIMULA version number, e.g_. SIMR41 and SIMR42 for version 4 of
SIMULA). SIMRn1 contains most of the routines, those of category a,
and the other high segment contains routines of category b and also
some routines of category a which are used by the category b
routines.
.s;The routines of category b are those concerned with program
initialization and file object creation. Communication between the
two high segments is via the transfer vector placed at the start of
the high segment and defined in the SIMRTS.MAC module via the
RTSYMBOLS macro. The transfer vector has the same size in both high
segments. The difference is in the contents of certain vector
elements. Each global subroutine in the high segment has a fixed
absolute position in the transfer vector. For those routines that are
present in the current high segment, the transfer vector contains a
JRST instruction to the actual entry point.
.p 0,1,4;A subroutine which is not present in the current high segment
is represented by a "PUSHJ XPDP,.OCSW" instruction in the transfer
vector, leading to the swapping routine.
.s;_.OCSW handles a call in the following way:
.br;All ac's are saved starting at YUUOAC(XLOW). If the call to the
eventual RTS routine was via a PUSHJ XPDP,xxxx instruction, the next
lower level of the stack should point to the instruction following the
PUSHJ. The alleged PUSHJ instruction is saved in the stack over the
GETSEG for some checking when the new high segment has been brought in.
A GETSEG UUO brings in the other high segment.
.s;Since, by design, the transfer vectors and the OCSW procedure are
identically placed in their respective high segments and exactly similar
except for a few locations, GETSEG will return to the next instruction,
which is now in the OTHER high segment (parallel worlds!!). Since
channel 0 is used for GETSEG, it must be restored if it was active for
TTY I/O before the swap, and the buffers must be relinked. The address
of the called entry in the transfer vector is retrieved from the return
address found on the stack.
.s;If the original call was a PUSHJ to this
address, OCSW must prepare to swap the other segment back in.
This is arranged by stacking as return address the address of L6,
where L6 is defined differently in the two high segments:
.br;1)####L6:#EXEC#.OCSW########2)########POPJ#XPDP,
.br;2)########POPJ#XPDP,#############L6:##EXEC#.OCSW
.br;OCSW calls the ultimately wanted routine by jumping to the
corresponding location in the current high segment, which should have
a JRST to the actual code.
If the subroutine returns to its caller, the return will be to L6
as defined above, causing a new swap and a final return to the caller,
since the location corresponding to L6 in one segment has a return
instruction (POPJ XPDP,) in the other segment.
.p 0,2,10
III.1.6.2.3####RTS error handling
.s;A Run Time System error is normally reported via the RTSERR UUO
as defined in the UNIVERSAL parameter file SIMMCR.
.br;Some errors are detected by other means, e.g_. arithmetic traps
(III.1.6.2.4), errors in routines from FORLIB (III.1.6.2.3.4),
errors during program initialization (III.1.6.2.3.5 and III.1.7.1).
.br;Some errors are left for the monitor to handle, such as failure to
find a high segment via GETSEG (a HALT instruction sends control to
the monitor), push down list overflow, exceeded time limit etc.
.br;In some situations which should not occur in a valid SIMULA
program + RTS, a RFAI UUO (defined in SIMMCR) is issued.
.p 0,2,10;III.1.6.2.3.1##"Normal" RTS errors in SIMULA (RTSERR)
.s;An error in a running SIMULA program is usually reported by
execution of the RTSERR UUO. The error number is indicated by the
address field of the UUO (no index, no indirect address).
An alternative way of indicating the error would be to use an ordinary
subroutine linkage instruction such as PUSHJ, JSR or an XCT applied
to some globally accessible location. The disadvantage is that another
word would be needed to specify the particular error, and since the
number of possible errors is rather large, this approach was rejected.
The main disadvantage with the UUO approach is a possible clash with
other uses of the same UUO, but since the SIMULA RTS must remain in
control in any case, that disadvantage is not felt to be great enough.
.s;The error UUO and other UUO's are handled initially by the RTS
routine OCUU.
.br;OCUU saves all ac's starting at YUUOAC(XLOW).
.br;The UUO opcode is checked and the following happens if the RTSERR
UUO is recognized:
.br;The error number is placed in YDSENR(XLOW) and the address of
the error UUO is placed in YDSEAD(XLOW). If the error did not occur
in compiled code (RTS stack not empty), a corresponding code address
is computed from the bottom element of the stack, and that address
is placed in YDSEAD instead.
.br;If SIMDDT is not already in core, OCLD attempts to load it to
the storage pool, possibly after expanding core or collection of
garbage records.
.br;If SIMDDT cannot be loaded, the error number is reported by an
inline error message and the program is terminated via OCEP (closing
files), otherwise SIMDDT gets control via its DSINE entry point,
writes the error message, identifies the source line and module,
then expects commands from the user. If SIMDDT returns (via an EXIT
command), the program is terminated (closing files etc.).
.s;For some errors, continuation is possible. Those errors are
distinguished by having a non-zero ac field in the error UUO.
The SIMDDT PROCEED command can be used to continue execution.
Error recovery is either automatic (standard action) or a new
text or integer value is requested to replace faulty data.
Extensions of this scheme are possible.
.p 0,2,10;III.1.6.2.3.2##RFAI UUO
.s;The RFAI UUO is used to signal an unusual situation in the RTS;
a situation which should not occur in a correct RTS, operating with
a correct SIMULA program (without any modules written in other languages
such as MACRO-10 or FORTRAN). An RTSERR UUO with a special error number
is simulated when the RFAI UUO is interpreted. The ensuing treatment
is described above.
.p 0,2,5;III.1.6.2.3.3##Illegal UUO
.s;If OCUU encounters an unrecognized UUO code, an error message is
again generated as for RFAI.
.p 0,2,10;III.1.6.2.3.4##Errors in FORTRAN library routines
.s;In library routines taken from the FORTRAN library (FORLIB), e.g_.
SQRT, errors are reported by the following instruction sequence:
.br;# XCT error-class, FORER.
.br;# CAI error-type, address(severity-code)
.br;Here error-class may be ER%LIB or ER%APR, where the errors of
class ER%LIB usually have an associated ASCIZ error message string
starting at "address", and the ER%APR errors simulate arithmetic traps,
where address is where to go on recovery.
.s;An ER%LIB error is handled by first typing the routine name in
an informative message prefixed by "ZYQEIR" then typing the message
string at "address" prefixed by "ZYQFLE". An ordinary error message
(RTSERR QFORER) is then simulated, transferring control to OCUU.
.s;For an ER%APR error, error-type specifies which trap
(integer overflow, floating point underflow, divide check, etc) should
be simulated. Error-type is translated to the equivalent SIMULA error
number, and an ordinary error message is again simulated.
.p 0,2,10;III.1.6.2.3.5##Errors during program initialization
.s;Some errors are detected in situations where SIMDDT cannot be invoked
because the necessary RTS environment is not yet set up. This occurs
e.g_. if the core size is too small to allow program execution, if
the high segment cannot be loaded, or if SIMDDT cannot be loaded.
Any error during the initial setup of SYSIN, SYSOUT etc in OCIN is
reported directly to the terminal without SIMDDT intervention.
.p 0,2,10;III.1.6.2.3.6##Errors diagnosed by the monitor
.s;Some errors cannot be handled easily within the RTS. Examples are
push down list overflow (presently, the stack is needed for error
handling), time limit exceeded.
.s;Attempted execution of illegal or privileged instructions,
execution of a HALT etc, will also lead to an error on monitor level,
as will some errors in connection with file handling.
See Monitor Calls (in the Software Notebooks).
.p 0,4,15;III.1.6.2.4##SIMDDT - RTS interface
.s;SIMDDT, the interactive SIMULA debugger, is described elsewhere in this document
(section IV). The RTS is responsible for bringing SIMDDT into core and calling
it in the following situations:
.br;- At program start in response to a DEBUG or REENTER command.
.br;- At program start to remove breakpoints if loaded earlier.
.br;- At a breakpoint inserted by SIMDDT.
.br;- Via REENTER after _^C interrupt.
.br;- When a RTS error has occurred.
.br;- After program exit in response to a CONTINUE or REENTER command.
.p 0,2,10;III.1.6.2.4.1##SIMDDT at program start
.s;SIMDDT will be called (entry point DSINI) before the start of the SIMULA
statements, but after RTS initialization, if execution starts at the entry
point .OCRE0 of the RTS. This is achieved via the DEBUG command or via the
REENTER command after a LOAD or GET or after program exit.
See III.1.6.2 (REENTER command).
When invoked in this way, SIMDDT stays in core during program execution and
also over a new START or REENTER command issued after program
exit. A subsequent START command will thus have the same effect as REENTER, i e
SIMDDT cannot be bypassed once loaded at program start.
After possibly setting some breakpoints, the user starts normal program
execution with a PROCEED command.
.p 0,2,10;III.1.6.2.4.2##Removing breakpoints on restart
.s;If the program is restarted after successful execution or after an error,
SIMDDT may have left breakpoints lying around if invoked dynamically via
_^C-REENTER. Any such breakpoints are removed by calling
OCSPDR at the start of OCSP or .OCRE0. OCSPDR determines if SIMDDT is in core
by checking the right half of .JBOPS, and if that is non-zero, the global
cell YDSBAS based on the address in .JBOPS. If YDSBAS is non-zero, it is
assumed to be the base of SIMDDT, which is called at the entry point DSINS
to remove the breakpoints specified in its internal table.
.p 0,2,10;III.1.6.2.4.3##Invoking SIMDDT at breakpoints
.s;When a breakpoint UUO is recognized by the UUO handler OCUU, SIMDDT is invoked
at its entry point DSINR, provided SIMDDT is available. Otherwise, an illegal
UUO is indicated.
.p 0,2,6;III.1.6.2.4.4##Invoking SIMDDT by _^C-REENTER
.s;
SIMDDT may be invoked at almost any program point by stopping the program with
one or two _^C commands, then issuing a REENTER command.
See III.1.6.1.2.3.
.p 0,4,6;III.1.6.3##SIMULA constructs for execution control - RTS support routines
.s;The facilities for execution control covered by section III.1.6.2 have no direct
counterparts in the SIMULA language definition. The present section describes
those execution control features defined in SIMULA
which are not implemented by compiled code.
.p 0,2,10;III.1.6.3.1##The state of execution of a block
.s;The state of execution of, e g, a procedure block, is defined by
the values of some items of information:
.br;- The values of all variables in the program.
.br;- The current program counter PC, also called program sequence control, PSC.
.br;- The current block CB, i e the nearest block with a display.
The accumulator designated XCB always points to the current block.
.br;- Possibly, the current subblock, implicitly defined by PSC and CB.
.br;- The reactivation point RP, consisting of a reactivation address RA
and a block instance address RBI, reactivation block instance.
In the RTS, RP=(RBI,RA) is represented in the display record by
(ZDRZBI,ZDRARE).
.br;- The state is also characterized by the Boolean conditions detached and terminated.
A block which is not detached is said to be attached, and a block
which is not terminated is said to be active.
.br;- In a wider sense, the state of a block is also affected by the
program environment, e g the state of I/O devices, other programs on the
machine including the monitor. These effects are not considered in this section.
.p 0,2,10;III.1.6.3.2##Procedure call, procedure exit
.s;A procedure is invoked via a procedure statement or a function designator.
Some procedures are implemented as inline coding sequences.
These are not considered here. Other procedures are implemented as standard
procedures which give a well-defined result and return to the caller without
essentially affecting the program state except for changing some variables.
These procedures are described in the sections on I/O handling, TEXT handling
and standard procedures respectively.
.p 0,2,10;III.1.6.3.2.1##Calling a SIMULA procedure
.s;A SIMULA procedure is set up by one of the subroutines CSSN or CSSW.
CSSN is used when setting up a "normal" procedure, i e a procedure which is
declared and statically visible from the point of call, and which is not
an attribute of an inspected class or a match for a virtual procedure.
Calling sequence:
.br;.nofill
.left margin 16
<save ac's>
.br;MOVEI XSAC,prototype address
PUSHJ XPDP,CSSN
.br;<parameter transmission>
.left margin 8
.fill
The sequence <save ac's> is only needed if intermediate results exist in some
accumulators (and possibly pseudo ac locations .YXAi). The sequence is:
.br;########PUSHJ###XPDP,CSSA
.br;########XWD#####n,admap
.br;where n is the number of intermediate results starting by XWAC1 and admap
is the address of a bit map specifying which ac's have object pointers
(see III.1.5.2).
CSSN allocates the procedure block and its attached display record (ZDR)
by calling SADB, then copies most of the display vector from that of CB.
RP is set to (CB,return address).
The ZAC address, if any, is copied by SADB to the display vector
from YCSZAC which is then reset to zero.
If the procedure has no parameters, CSEN is entered directly instead of
returning to the calling program.
On exit from CSSN, XRAC=XWAC1 holds the address of the procedure block.
.p 0,1,5;The calling sequence to CSSW is
.br;########[evaluate ZDP of procedure to top ac's]
.br;########PUSHJ###XPDP,CSSW
.br;########XWD#####n,admap
.br;########<parameter transmission>
.br;The top ac's are XWAC1+n and XWAC2+n. If n is zero, admap is also zero,
signifying that no intermediate results exist in ac's.
CSSW allocates a procedure block and display vector via SADB like
CSSN. The ZDP information is saved in YOBJAD(XLOW)
over a possible garbage collection and relocated there if necessary.
The innermost level (furthest from the block in the display) is the
address of the procedure block itself, put there by SADB.
The next to innermost level in the display is taken from ZDPEBI of the input
dynamic procedure address, and the rest of the display is copied from
the display of ZDPZBI.
.s;III.1.6.3.2.1.1##Code for parameter transmission
.s;The sequence <parameter transmission> is only needed if the procedure has
parameters and has the following form:
.br;########<parameter coding>
.br;########PUSHJ###XPDP,CSEN#######;Enter procedure coding
.br;If the procedure has parameters, CSSN or CSSW returns to SIMULA code for
evaluation and transmission of the actual parameters.
The coding sequence, <parameter coding>,
compiled to transmit the parameters must preserve
the value of XWAC1 by using XWAC2 as top ac and causing a ZAC object
to be created whenever the garbage collector may be called.
The <parameter coding> consists of coding to evaluate and transmit a
representation of each actual parameter.
The code generated for a parameter normally consists of a
straightforward expression evaluation to one or two ac's and a store
operation which places the value in the formal location. For name parameters,
a parameter descriptor (ZFL instance) is computed and stored, and any thunk
coding is compiled inline and bypassed with a JSP instruction.
For parameters by reference or by value, the normal representation is used
for variables and expressions.
For procedure, label or switch parameters, a dynamic representation is
computed (ZDP, ZDL or ZDS instance) and stored. For a TEXT parameter by
value, the system routine COPY is called, and for an ARRAY by value,
CSCA is called.
When calling a formal, virtual or external NOCHECK MACRO-10 procedure,
however, the parameter transmission sequence following the PUSHJ to CSSW
has a special form involving
a call on the RTS routine PHPT. See parameter handling.
CSEN is called directly from CSSN or CSSW if no parameters exist, otherwise
it is called after parameter transmission to start the
operations of the procedure.
.s;III.1.6.3.2.1.2##Starting the actions of a SIMULA procedure
.s;CSEN changes XCB and jumps to the declaration coding of the procedure.
A procedure, unlike a class object, can never be detached. When it terminates
via CSEP for a function procedure or CSES for a pure procedure, its data area
can be reclaimed directly or by subsequent garbage collection.
.p 0,2,10;III.1.6.3.2.2##Calling FORTRAN and MACRO-10 procedures
.s;OPTIONS(/E:FORTRAN (or F40),entrypoint);
.br;The coding sequence for calling an external FORTRAN procedure is
essentially the same as that for any SIMULA procedure,
with the restriction that REF, LABEL, SWITCH, and PROCEDURE
parameters are not allowed. A procedure
block is allocated in accordance with a special prototype
generated from the specification coding for the procedure.
The special routine PHFO serves as intermediary between the SIMULA-generated
procedure block and the FORTRAN code. See parameter handling.
When the FORTRAN code is active, XCB is saved in the global cell YFOXCB.
.s;OPTIONS(/E:CODE,NOCHECK,entrypoint);
.br;A NOCHECK MACRO-10 procedure is handled like a formal or virtual SIMULA
procedure, i e by a special calling sequence to CSSW, including a call on PHPT.
See parameter handling.
Strict rules must be followed by the MACRO-10 procedure in order not to
compromise the security of the SIMULA program.
.s;OPTIONS(/E:CODE,entrypoint);
.br;A MACRO-10 procedure with specified parameters is handled exactly as a
SIMULA procedure.
.s;OPTIONS(/E:QUICK [,NOCHECK] ,entrypoint);
.br;A MACRO-10 procedure which is specified as "QUICK" (not "CODE") takes
all its parameters in ac's and is called by a PUSHJ XPDP,entrypoint
instruction. The calling sequence is the same as for e.g_. MAIN,
STRIP. The RTS is not involved, only compiled code. See SIMLH2.MAN
Chapter 7 and appendix E.
.p 0,2,10;III.1.6.3.2.3##Procedure exit
.s;The exit from a procedure is via the final statement or possibly via a GOTO
statement. A function procedure exits via CSEP, which restores any intermediate
results and puts the function value on top. The presence of intermediate
results is signalled by the ZDNACS bit in the object, and the saved values
are retrieved via the ZDRZAC pointer in the display record.
Two ac's are always transmitted
to avoid testing for result type in CSEP. The compiled code knows which
ac(s) to use according to the stack discipline for register allocation.
A pure procedure exits via CSES, which is similar to CSEP except that
no function value is returned.
The procedure block may be deallocated if no block which
is still referenceable was allocated on top of the procedure block.
.index Block deallocation???????
A procedure always returns to the environment given by (ZDRZBI,ZDRARE),
the dynamic link, except for a direct or indirect GOTO statement.
.p 0,2,10;III.1.6.3.3##Creation of a class object
.s;A class object is generated by an "object generator" - NEW <class-id>
[<parameter list>].
An object generator is very similar to a function designator and may also
serve like a procedure statement if the value is discarded.
Since name mode parameters are not allowed, the calling sequence is simpler
than a general procedure call.
Calling sequence:
.br;########<save ac's>
.br;########PUSHJ###XPDP,CPNE
.br;########XWD#####m,prototype address
.br;########<parameter transmission>
.br;where m=offset in display of the block whose display should be copied.
Usually, the display of XCB can be copied, but in some cases of inspection
statements all relevant levels may not be correct.
CPNE works in a fashion very similar to CSSN. If the class has local classes
declared (ZCPKDP bit set in prototype), the ZDNKDP bit in the block must be set
to prevent the display from being deallocated
when the class object is terminated.
The display may be needed when calling a procedure attribute of one of the local
classes, which may again be terminated and without a display.
Any REF or ARRAY locations on the current prefix level of the object are
initialized to NONE via SADB on allocation.
CPIN is called to initialize REF and ARRAY cells in the outer prefix levels.
If parameters exist, control goes back to SIMULA code for parameter
transmission, otherwise CSEN is called directly.
Parameter transmission is done as for a procedure but is simpler since
only value and reference mode parameters are possible and label, switch
and procedure parameters are not allowed. See III.1.6.3.2.1.1.
CSEN enters the declaration coding of the outermost prefix. CPCD transfers
control to the declaration coding at the next inner level or the statements
at the outermost level.
CPCI transfers control to the statements of the next inner prefix level
at an explicit or implicit INNER statement. At the end of the statements at
the innermost level without an explicit INNER,
CPE0 is called to terminate and detach the class object.
If an explicit inner exists for some prefix, compiled code effects the
transfer either directly or via ZCPIEA of the prefix, if the code address
is not available at compile time (external or system class).
Control returns with an object reference in the top ac either when the class
terminates or when an explicit or implicit detach is performed. See below
(quasi-parallel sequencing).
.p 0,2,10;III.1.6.3.4##Entering a prefixed block
.s;A prefixed block is set up in a way very similar to a class object.
Calling sequence:
.br;########MOVEI###XSAC,prototype address
.br;########PUSHJ###XPDP,CPSP
.br;########<parameter transmission>
.br;The display is copied from the statically enclosing block, adding the address
of the prefixed block itself at the innermost level.
Initialization is done as for a class object.
In order to implement quasi-parallel sequencing as described by CBL 9.2,
an enclosing detached block must be linked dynamically to this prefixed block.
This is done by setting RP of the nearest enclosing detached block to
(address of this prefixed block, 0).
If this is the outermost prefixed block, the dynamic link is set to point to
itself.
After any parameter transmission, which is similar to that for a class,
CSEN transfers control to the declaration
coding at the outermost prefix level as for a class. Instead of the last call
to CPCD, however, CPPD is called to transfer control to the statements
of the outermost prefix. The block terminates via compound tails of prefixes and
ultimately via CPE0, which restores
the XCB of the static environment before exiting to the instruction after
the prefixed block.
.p 0,2,6;III.1.6.3.5##Implementation of quasi-parallel systems
.s;Quasi-parallel systems (QPS) are described in CBL 9.2.
In the present section the actual implementation of a QPS control structure is
outlined.
.p 0,2,10;III.1.6.3.5.1##Information needed to define a QPS
.s;A QPS consists of
.br;a) A prefixed block, called the main program of the QPS.
.br;b) Any enclosed detached class objects or prefixed blocks.
.br;c) Any other blocks enclosed by the main program.
.br;The COMPONENTS of the QPS are the detached instances.
Blocks of category (c) are relevant only to define "enclosed",
which is interpreted according to CBL 9.1, q v.
The QPS structure is defined by the reactivation points of its components.
The reactivation point RP = (RBI,RA) of a component is interpreted in the
following way:
.br;If RA is zero, RBI points to an enclosed prefixed block which contains the
actual reactivation point. If RA is non-zero, it points to the program address
where execution should resume, with XCB=RBI.
.s;The basic means for controlling the execution of a QPS are provided by the
system procedures DETACH and RESUME, CBL 9.2.
The procedure CALL has been implemented to simplify programming of
coroutines.
.p 0,2,10;III.1.6.3.5.2##DETACH statement
.s;The detach statement works according to CBL 9.2.1.
The effects described there are implemented by the CPDT procedure
in the following way:
.s;a) The smallest operating block instance, X, is attached, either because it
has just been generated via NEW or because it has been reattached via CALL.
.br;Actions: RP := (X,RA), where RA is program point after call on detach.
ZDNDET is set, and control returns to old RP (calling block,return address).
.s;b) X is already detached.
.br;Actions: RP := (X,RA) as above.
The main program of the innermost enclosing QPS is then found by following
static links for detached objects, dynamic links for attached objects, until
a prefixed block is found. The static link SL of a block B
is the block address found at
offset B.ZBIZPR.ZCPSBL relative to B.
RP = (BI,RA) of the main program is found. If the program address RA is zero,
the reactivation point is inside BI (an enclosed prefixed block), and
RP := BI.RP, whose program address RA is then examined, etc until
a non-zero program address is found. This is where execution resumes, with XCB
set to RP.BI (ZDRZBI).
.p 0,2,10;III.1.6.3.5.3##RESUME statement
.s;The resume procedure has the effects described in CBL 9.2.2.
The RTS routine CPRS implements resume.
Assume that the detached object to be resumed is Y, component of a QPS called S.
Resume is called inside a currently operating component X of the same QPS.
.s;Actions: X, the nearest detached block instance, is found by dynamic links
starting from XCB. RTS error (RESUME operating) if X=Y.
X.RP := (CB,RA), where CB is current block, pointed to by XCB and
RA = address after resume statement.
To check for errors, all enclosing detached instances are found by
following static and dynamic chains from X as in the detach procedure
(the operating chain is followed) until the outermost block is reached.
If any of the encountered blocks is Y, we have an error.
Starting from Y, find actual reactivation point by following RP = (BI,RA)
until RA is non-zero. Resume execution at that RP.
.p 0,2,10;III.1.6.3.5.4##CALL statement
.s;The CALL procedure (CPCA) is not part of the Common Base but is a recommended
extension.
CALL essentially reverses the effect of DETACH, as follows:
.br;Assume the statement CALL(Y), executed within a QPS component X.
X and Y are components of the same QPS S.
Y may not be attached, terminated or operating.
X is found starting from XCB, following RP until a detached block is found.
Error if X=Y. Copy Y.RP to X.RP to set correct reactivation point.
As in CPRS, all detached blocks enclosing X are found via
static and dynamic links and checked against Y (error if identical).
Attach Y to caller by setting off the ZDNDET bit and setting
RP = (CB,RA), where CB is
current block, RA is program address after CALL statement.
The reactivation point of Y is found as in CPRS and execution resumed there.
.p 0,2,10;III.1.6.3.5.4##Object termination
.s;Ignoring any compound tails after explicit INNER statements,
the effect of passing through the final END of an object is to detach
and terminate the object. CPE0 is called.
For a class object, CPE0 sets the ZDNTER bit and transfers control to CPDT
(see detach). The same is done for a prefixed block, but since detach has no
effect, control is returned to CPE0 which restores XCB and transfers control
to the statement after the prefixed block.
.p 0,2,4;III.1.6.3.5.5##GOTO statement leading out of an object
.s;See CBL 9.2.4.
.p 0,2,6;III.1.6.3.6##INNER statement
.s;The INNER statement is implemented by the CPCI routine, which transfers control
to the statement coding at the next inner prefix level. Has no effect if there
is no inner level.
.p 0,2,10;III.1.6.3.7##GOTO statements, switches
.s;GOTO statements are normally handled by compiled code. If the target of the
GOTO implies change of current block (CB), CSGO is called, however.
The parameter to CSGO is a dynamic label (ZDL instance).
CSGO follows the operating chain from the current block until the target block
is found or the outermost block is reached.
Class objects passed on the way are terminated and subblock addresses
eliminated from displays when passing.
Since the target must be operating, a
GOTO out of a detached object is allowed only if it leads out of the nearest
enclosing prefixed block, and a transfer to a connected label will be allowed
if and only if the inspected block is operating.
A SWITCH element is evaluated by CSSC, which takes a dynamic switch address
(ZDS instance) and a switch index as parameters and returns a dynamic label
address (ZDL).
If the switch element is a statically visible label, the static description is
translated to the dynamic form.
If the switch element is a designational expression, a dummy procedure block
is set up and the switch thunk is entered, to return later via CSES.
CSES works like CSEP except that the dynamic label is already in
the top ac's.
.p 0,2,10;III.1.6.4##Execution control by compiled code
.s;Apart from calling sequences to RTS routines,
the principal means of execution control in compiled code are
conditional statements and expressions, WHILE loops, FOR loops and
simple GOTO statements.
Only FOR loops are interesting from the RTS point of view, because
the return address of a FOR loop is stored in the display vector. One word
is required in the display vector for each FOR loop nesting level in the
current block.
.pg;III.1.7##I/O handling
.br;#########------------
.p 0,2,10;I/O handling in a SIMULA program covers the following basic areas:
.s;BASICIO functions, i e setting up and opening SYSIN and SYSOUT.
In DECsystem10 SIMULA, file specifications may be read and
runtime switches interpreted.
.s;Creation and setup of file objects of the various file classes, allocation
and handling of buffers.
.s;Implementation of the basic I/O functions OPEN, CLOSE, INIMAGE, (BREAK)OUTIMAGE,
LASTITEM, LOCATE, ENDFILE.
.s;I/O editing functions: INCHAR, OUTCHAR, ININT etc.
.p 0,4,10;III.1.7.1##BASICIO functions
.s;The functions related to the definition of BASICIO in CBL 11 are mostly
implemented in the OCIN module of the RTS. Pointers to the standard
file objects SYSIN and SYSOUT are placed in the static lowseg area at offsets
YSYSIN and YSYSOUT respectively. FILE prototypes are placed in the
transfer vector (SIMRTS module), whereas declaration and statement coding and
symbol tables are placed in the IO module.
A special naming scheme is adopted for quantities relating to file subclasses:
.br;Each class is denoted by 4 letters: IOFI (FILE), IOIN (infile), IOOU (outfile),
IOPF (printfile) and IODF (directfile).
SIMRPA defines a macro for each prototype. If the class code is xxxx (=IOIN etc),
then .xxxx is the global name (prototype base address), xxxx%S is the address
of the initial statements, xxxx%D is the start of declaration coding, D%xxxx is the
name of the macro, xxxx%I is the first address after inner in the statement part,
xxxx%M is the block map address, and xxxx%Y is the address of the symbol table.
.s;Before allocating SYSIN and SYSOUT, a possible specification file defined via
a compile-time /R switch is read. Encountered switches are acted upon, buffer
space is allocated, and a specification table for files is built.
This table will be consulted whenever a file subclass is created.
.s;When any specification file has been processed, _.SAGI is called to
initialize the storage allocation system, leaving the buffers and specification
tables below the bottom of the storage pool.
.s;Next, the SYSIN infile object is set up.
Normally, SYSIN and SYSOUT will both be the user's TTY, but the specification
file may define SYSIN and/or SYSOUT differently. In the absence of
specifications for SYSIN or SYSOUT, the logical name SYSIN or SYSOUT is checked
to find out if it has been assigned as a device name. If so, the file named
SYSIN or SYSOUT on that device is used, otherwise the TTY is used.
See OCIN for complete algorithm.
SYSIN.open(blanks(n)) is performed, where n=80 or the width of the TTY line in
the case when SYSIN is a TTY.
.s;The SYSOUT printfile object is then set up, and LINELENGTH is determined
as 132 or the width of the TTY line if applicable. The actual execution of
SYSOUT.open(blanks(LINELENGTH)) is left to OCEI which is then entered.
.s;The final actions of BASICIO, SYSIN.close and SYSOUT.close, are performed
at program end via OCEP.
.p 0,2,10;III.1.7.2##File object generation, file setup, buffer allocation
.s;The calling sequence for a file object generation is similar to that
for a SIMULA-coded class, for example:
.br;.left margin 16
PUSHJ###XPDP,CPNE#######;NEW infile("INPUT")
.br;XWD#####-n,IOIN
.br;DMOVE###XWAC2,[XWD#0,-2+[ASCII/INPUT/]
.br;###############XWD#5,0]
.br;PUSHJ###XPDP,TXCY
.br;XWD#####1,[1B0]
.br;DMOVEM##XWAC2,OFFSET(ZFISPC)(XWAC1)
.br;PUSHJ###XPDP,CSEN
.br;.left margin 8
This calling sequence, when executed, causes the following actions:
.br;- The file object is created via CPNE and SADB. The display is copied
from the block referenced at offset -n in the display of the current block.
.br;- CSEN changes XCB and sets the return information, then transfers
control to the declarations of the outermost prefix level, IOFI%D.
Neither IOFI%D nor IOIN%D, which is entered next, contains any real
declaration coding, only calls CPCD. The statement coding part of IOFI,
IOFI%S, which is then entered, is also empty, transferring control to
IOIN%S via CPCI (inner).
At IOIN%S, the real initial actions for the infile start by setting flags
and clearing certain switches, then calling the SETUPFILE routine of the
IONF module
.s;- SETUPFILE uses the information in ZFISPC
(copy of text parameter to NEW ...)
and any specification table created at BASICIO initialization (see above) to
determine the characteristics of the file which is to be input and/or
output. If sufficient information is not available or a dialogue is expected,
the user is interrogated for the missing information.
A free software channel is claimed and OPENed.
Device and file name, number and size of buffers, etc is determined,
if necessary by consulting the user.
CREATEFILE allocates and links the buffers, performs new OPEN on the same
channel as before, and performs LOOKUP and/or ENTER.
.p 0,2,10;III.1.7.3##Other basic I/O functions
.s;Apart from the I/O functions mentioned above, the following procedures contained
in the IO module are
essential: OPEN, CLOSE, INIMAGE, (BREAK)OUTIMAGE, LASTITEM, LOCATE, ENDFILE.
The remaining procedures are merely applications of text handling procedures.
Most of the procedures conform to the definitions in CBL chapter 11.
Special information:
.s;- OPEN (IOOP): Special actions must be done via the routine REOPEN in IONF
if the file was once opened and then closed.
Buffers must be allocated and chained, information fetched from
the specification table, an OPEN UUO must be performed etc.
These things are normally done at file creation (see above), and the SIMULA
procedure OPEN simply does what CBL specifies.
For DIRECTFILES, however, OPEN computes file record length
from image.length, allowing for <CR><LF> and rounding up
to a multiple of 5 (word alignment), and computes
max LOCATION from file size and record length.
.s;- CLOSE (IOCL), in addition to doing what CBL prescribes, also
outputs the last image for an outfile, provided pos > 1.
The software channel is released and YIOCHTB updated.
If the channel table entry refers to two file objects
representing a bi-directional device such as a terminal,
the channel is not released until the last of the two
files is closed.
.s;- INIMAGE (IOIG) has the following special properties:
Carriage returns and null characters are ignored. Line feed is taken
as end of image but is not transmitted to the image. Form feed, vertical tab
and altmode (octal 14, 13, 33) are taken as image delimiters and are also
taken as part of the image.
If inimage tries to read past the end of the file or reads an unwritten record
inside a directfile, a special end-of-file record starting with /* and padded
with blanks is transmitted to the image.
.s;- OUTIMAGE (IOOG). Only image.strip is moved to the buffer to save
file space and time. Carriage-return is tagged on to the end.
If the buffer is output immediately, as for a TTY file, a line feed
is added before output. Pending line feeds from the last line, if
any, are output to the buffer before the current line is output.
This is because of our interpretation of SPACING for PRINTFILES.
.s;- BREAKOUTIMAGE (IOBO). Works as OUTIMAGE, except that no CR-LF is added.
To enable spaces at the end to be significant, image is not stripped
shorter than image.pos-1.
.s;- LASTITEM (IOLI) works according to CBL, except that horizontal tab (octal 11)
is treated like the space character.
.s;- ENDFILE is implemented as an ordinary variable in the infile or directfile
object. ENDFILE is set to -1 (TRUE) when inimage finds that
the file is exhausted, i.e_. when the last character in the file
was read by the previous inimage call.
.s;- LOCATE merely checks if the argument is in range and if so, sets off
endfile flags and sets a new LOC value = argument.
No operations directed to an external device are performed here.
.s;Special considerations for directfile:
.s;- The images in a directfile are all of the same length, given by the
size of the text given as IMAGE when calling OPEN.
Images are buffered within disk buffers (only DSK
allowed as device), and the disk buffers are held in core as long as
possible to minimize actual I/O. When inimage or outimage is performed on
a disk block not currently in core, the block is read first if it exists.
If no outimage was performed on a buffer, it is not rewritten on disk when
replaced in core.
Close does not imply that the last image is written as for outfiles.
.p 0,2,10;III.1.7.4##Editing procedures for input/output.
.s;The editing procedures, e g INCHAR, ININT, OUTFRAC, are defined in terms of
the corresponding text attributes. The definitions given in CBL
usually cover those functions sufficiently.
The "inaccessible" function FIELD of CBL 11.3.1 is implemented as .IOFD.
Deviations from CBL:
.br;INCHAR (IOIC) and INTEXT (IOIT) test for ENDFILE before calling inimage.
In this way characters from the end-of-file image can be read.
.pg;III.1.8##Text handling
.br;#########-------------
.s;Text handling is performed by the modules TX and TXBL which contain a) RTS
procedures and functions corresponding to the text attributes found in section
10 of CBL, b) the non-local procedures BLANKS and COPY. Text relations and
text value assignment are also handled by special procedures. A routine for
creating a temporary text variable (ZTT) is also part of the TX module.
Some attributes are implemented by inline coding.
The representation of text information is described in I.5, data structures,
and LH2 appendix B.
.p 0,2,10;III.1.8.1##Text attributes
.s;Most text attributes are implemented by straightforward translation of the
CBL definition. Some implementation dependent details are outlined below.
.br;Pos is ZTVCP+1, i e pos=1 implies ZTVCP=0 which is convenient since it
allows NOTEXT to be represented by two zero words.
The attributes LENGTH, POS and MORE are implemented by inline code.
The function attributes MAIN (TXMN), GETCHAR (TXGC), GETINT (TXGI),
GETREAL (TXGR) and GETFRAC (TXGF) are called as follows:
.br;########MOVEI###XTAC,Xtop
.br;########PUSHJ###XPDP,function
.br;where Xtop and Xtop+1 contain the text reference on entry and the returned
function value on return.
The other attributes: SETPOS (TXSE), PUTCHAR (TXPC), PUTINT (TXPI),
PUTREAL (TXPR), PUTFIX (TXPX) and PUTFRAC (TXPF) are called with the text
reference in XWAC1 and XWAC2, other parameters in XWAC3, XWAC4, etc.
The additional procedure LOWTEN (TXLT) is called in the same way.
.p 0,2,10;III.1.8.2##Other text procedures
.s;Texts are generated by COPY (TXCY), BLANKS (TXBL) and by INTEXT (IOIT), which
is part of the I/O sub-system (III.1.7).
Text assignment of the form T1 := T2 is performed by TXVA, and comparison
of text values, e g T1 < T2, is performed by TXRE. TXRE and TXVA are
called with XTAC pointing to the first ac of the 4 consecutive ac's
that contain the texts T1 and T2 respectively.
The relation between T1 and T2 is coded in the result of TXRE as -1, 0 or +1.
TXCY is called with an inline parameter specifying any intermediate results:
.br;########PUSHJ###XPDP,TXCY
.br;########XWD###n,admap
.br;Thus an accumulator stack may have to be saved over storage allocation.
The parameter to TXCY is saved in YTXZTV over a possible garbage collection.
TXBL is called like TXCY.
The text procedure TXDA, called like TXCY, in certain cases
creates a ZTT object containing a dummy text variable which stands for a
text expression. The output of TXDA is the dynamic address of this text
variable i e [XWD 1,ZTT object address] and also its absolute address in the
next ac.
.pg;III.1.9##Parameter handling
.br;#########------------------
.s;Since value and reference mode parameters to ordinary procedures, classes
and prefixed blocks are handled entirely by compiled code, parameter handling
in the RTS is limited to unusual circumstances - treatment of name mode
parameters, transmission of parameters to formal, virtual, FORTRAN and NOCHECK
MACRO-10 procedures.
.p 0,2,10;III.1.9.1##Access to name mode parameters
.s;Inside a procedure, a name parameter is accessed in different modes and
contexts. The PH module contains run-time procedures for handling
name mode parameters, calling thunks, etc.
.p 0,2,10;III.1.9.1.1##The value of a name mode parameter
.s;If the value of the parameter is required, PHFV is usually called.
PHFV handles parameters of kind simple, i e not array, procedure or switch.
Label parameters are handled by PHFM, and text parameters are handled by PHFV
only in contexts where the text descriptor will not be modified, as in
T.strip.
.s;PHFV starts by calling PHINIT (see below).
The parameter does not have a thunk if the actual parameter is a simple
variable or constant. The absence of a thunk is signalled by the ZFLNTH bit
of the formal location. In this case, the value of the parameter is loaded
directly from the ZFL instance, from a block instance or from the literal pool.
See ZFL format. In other cases, the parameter must be computed by a thunk.
This is done via ENTERTHUNK, PROCVALUE (if actual parameter is a parameter-less
procedure) and THUNKRETURN. Since calling the thunk implies loss of control
(the only link back to PHFV is via the return address for ENTERTHUNK), care
must be taken to call ENTERTHUNK in such a way as to retain all useful
information.
If the thunk returns a dynamic address, i e the thunk is not a value type
thunk (ZFLVTD bit), the value must be retrieved via the dynamic address.
Conversion of the value is done by PHCV (entry PHCV1) if necessary
(ZFLCNV bit).
PHEXIT returns to compiled code with the value in the proper ac(s).
.p 0,2,10;III.1.9.1.2##The dynamic address of a name mode parameter
.s;The dynamic address of a parameter X is needed
.s;a) when X is used on the left hand side of an assignment (X:=... or X:-...).
The actual parameter in this case may be a simple variable,
a value or reference mode parameter or an array component, possibly
accessed via remote referencing (<object expression>.<attribute>).
PHFA is used in this case, except for type TEXT, which is handled by PHFV
for value assignment and PHFT for text denotes (X:-...).
.s;PHFA works as follows:
PHINIT is called. Then, if the ZFL type for the parameter signifies a
value rather than a dynamic address, an error message is issued.
If no thunk exists, the dynamic address is computed from the ZFL, otherwise
the thunk is evaluated via ENTERTHUNK and THUNKRETURN.
If the parameter type is REF, the prototype pointer is taken from ZFLZQU
before returning to compiled code via PHEXIT.
.s;b) when X is specified PROCEDURE, ARRAY, SWITCH or LABEL.
PHFM is used in this case. PHFM works like PHFA except that no check for
illegal assignment need be made.
.s;c) when X is specified TEXT and used in a context where the text descriptor
may be affected as with text editing procedures like GETCHAR, PUTINT.
PHFT is used in these cases and for text denotes (X:-...).
.br;PHFT starts by PHINIT and works essentially like PHFA if the parameter
representation yielded via the ZFL or a thunk is a dynamic address.
In this case, the actual parameter, possibly accessed by
remote referencing, is a text variable (or parameter by
value or reference) or a text array component.
.br;If the actual parameter is a constant, a RTS error is signalled. This is so
if ZFLVTD=1 and no thunk has been compiled.
The remaining case is when the actual parameter is a text expression computed
by a thunk. Evaluation is done as in PHFV, but before returning to compiled
code, a dummy text variable must be created. TXDA creates a ZTT record with
a copy of the descriptor computed for the text expression. The dynamic
address of this dummy descriptor is returned from PHFT to compiled code.
.p 0,2,10;III.1.9.1.3##Assignment to formal parameters by name
.s;If X is a formal parameter by name, an assignment of the form X:=<rhs>
is carried out as follows: a) PHFA computes the dynamic address to one or
two ac's. These ac's are preserved over the execution of <rhs>.
.br;b) The assignment is carried out by PHFS. Assume that the dynamic address of
X is in XWACi (and possibly a qualification in XWACi+1), and the value of <rhs>
in XWACj (+XWACj+1). The calling sequence to PHFS is:
.br;########HLLZ####X0,X###;ZFL instance
.br;########PUSHJ###XPDP,PHFS
.br;########XWD#####XWACj,XWACi
.br;PHFS carries out qualification checking and possible conversion
before transmitting the value.
.p 0,2,10;III.1.9.1.4##Accumulator handling in PH procedures
.s;The procedures PHFA, PHFM, PHFT and PHFV interface with compiled code
via the two local subroutines PHINIT and PHEXIT.
The calling sequence to any of the procedures is
.br;########[dynamic address of X to XWAC1+n]
.br;########PUSHJ###XPDP,PHFy
.br;########XWD#####n,admap
.br;where y is A, M, T or V, n is the number of intermediate results and
admap is zero or the address of a relocation map for the n results.
PHINIT calls .CSSA. to save these results if n > 0, then puts
the dynamic address of X in XWAC1 (normal case).
The absolute address of X, the formal location or ZFL instance,
is loaded to XFAD, and the two ZFL words are loaded to XFL0 and XFL1.
After computing the parameter, possibly via a thunk, each routine
returns to compiled code via PHEXIT.
PHEXIT modifies the stack to skip the inline parameter. If intermediate results
were saved (YCSZAC(XLOW) nonzero), PHEXIT returns via .CSRA, otherwise directly.
.p 0,2,10;III.1.9.1.5##Thunk evaluation
.s;The execution of a thunk is started via ENTERTHUNK, which is called by
JSP XRET,PHET. Before executing the thunk code, information must be
placed in the thunk save area allocated in the display of the block
where the actual parameter belongs. The offset of the thunk save area (ZTS)
is found at the head of the thunk as shown by the following general coding
sequence compiled for a thunk:
.p 0,1,6;########XWD#####d,0
.br;########<code to compute the actual parameter>
.br;########MOVEI###XSAC,d(XCB)
.br;########JSP#####d+OFFSET(ZTSRAD)(XCB)
.s;where d is the offset of the thunk save area in relation to the end
of the display vector.
In the thunk save area are saved the dynamic address of the formal
location, the return address to object code (found in XPDP stack), the
return address to the present RTS routine (value of XRET), the acs
pointer (YCSZAC) and XCB. This information is sufficient to restore
the environment on return from the thunk, provided that the save area
itself can be found, which is guaranteed by having the thunk return its
address in XSAC as seen above. The new XCB is taken from ZFLZBI, and the
thunk is entered with the ZTS address in XSAC.
.s;In the rare case when the value of a procedure without parameters is
required, the thunk makes an intermediate return with the dynamic address
of the procedure in XWAC1 and XWAC2. PROCVALUE (called by JSP XRET,PHPV)
checks if the procedure should have parameters after all (error if so), and
then returns to the thunk with a new return address taken from XRET put
into the save area. X0 is stacked on entering PHPV so that SIMDDT can get
the correct code address for the error message.
.s;When the thunk returns to the RTS, which may be after execution of a substantial
amount of code including execution of the thunk itself recursively,
THUNKRETURN uses the information in the thunk save area based on XSAC to
restore the environment where the parameter was accessed. On exit from
THUNKRETURN (PHTR), XFAD points to the formal parameter X (ZFL instance),
the return to SIMULA code is stacked, YCSZAC points to any saved intermediate
results, XCB is restored, and XFL0 contains the first ZFL word.
.p 0,4,6;III.1.9.2##Parameter transmission to formal, virtual and NOCHECK MACRO-10
procedures
.s;The parameters to formal, virtual and NOCHECK MACRO-10 procedures
cannot be checked fully at compile time,
since the formal parameter specifications are not known.
Procedures in this category are called via CSSW. The computation
and transmission of actual parameters is carried out by the RTS procedure PHPT,
which has a very special calling sequence:
.lm 16;.nf;.p 0,1,24;
[compute dynamic address of formal or virtual
procedure to top ac's]
PUSHJ XPDP,CSSW
XWD n,admap
Z npar
PUSHJ XPDP,PHPT
[Z prototype address] ;Only for REF
.i -8;ZAP1: actual parameter descriptor (ZAP) for first param.
XWD d,ZAP2
[thunk for first parameter, if any thunk is needed]
[Z prototype address] ;Only for REF
.i -8;ZAP2: ZAP for second parameter
XWD d,ZAP3
[thunk for second parameter]
...
[Z prototype address] ;Only for REF
.i -8;ZAPn: ZAP for n'th parameter
XWD d,ZAPEND
[thunk for n'th parameter]
.i -8;ZAPEND: XWD 0,0
PUSHJ XPDP,CSEN ;enter procedure body
.f;.lm 8
.s;In the above sequence, d is the offset of the last word of the
thunk save area in the display vector, and npar is the number of actual
parameters.
CSSW is informed that PHPT is to be called by the [Z npar] instruction
placed as a second inline parameter to CSSW.
In any other situation involving CSSW, this instruction must be an executable
instruction. CSSW uses the absence of this information to give an error message
if the procedure expects parameters but has not been provided with an actual
parameter list.
.s;The number of actual parameters is checked against the number of formal
parameters, except for a NOCHECK MACRO-10 procedure (ZPCNCK bit set in the
prototype).
The parameters are treated in sequence. Information from the ZAP
or the ZFL in case of a passed-on name mode parameter is loaded
to standard registers. In the NOCHECK case, the formal parameter
type and kind are copied from those of the actual parameter.
The prototype of a NOCHECK MACRO-10 procedure has formal parameter descriptors
(ZFP) for 31 formal parameters all specified name. This means that
ZFL instances will be computed and stored for all parameters.
The actual parameter is first
checked for compatibility. If ZAPNTH is set, the ZAP specifies
the address or value of the quantity directly (generally, by
effective block level and offset). The quantity can be loaded,
possibly converted or qualification checked, and stored in the formal location.
Note that the qualification of an actual REF parameter is found in the
word preceding the ZAP. If the formal parameter is specified NAME, a
ZFL instance is computed and stored. A NOCHECK MACRO-10 procedure is treated
as having only name mode parameters. If a thunk has been
compiled, and the formal parameter is not of name mode, the
thunk must be evaluated to yield the value. This is done in the
same way as in PHFA or PHFV, with the additional requirement
that the current positions in the actual (ZAP) and formal (ZFP)
descriptor lists must be remembered. This is done by storing
the addresses of the ZAP and the ZFP descriptors in the formal
location until the value has been computed. Since the dynamic
address of the formal location is saved in the thunk save area,
the formal and actual descriptor positions can be recovered on
return from the thunk. The dummy ZAP of all zeros finishes the
parameter list.
.p 0,2,6;III.1.9.3##Parameter transmission to FORTRAN procedures
.s;A FORTRAN procedure is identified to the SIMULA system via the attribute
file produced by the SIMULA compiler when the interface code is compiled.
See LH2 appendix E.
The prototype of the procedure and its associated core map (ZMP)
contains information assuming a special block layout documented
in the PHFO module. The value of ZPCDEC is the address of PHFO, which
is thus entered after block allocation and parameter transfer, which
is done exactly as for a SIMULA procedure.
PHFO has the task of adapting the parameters to FORTRAN formats and
calling the actual FORTRAN procedure. If the returned parameter values
have to be converted, PHFO does this before returning to SIMULA code.
Both old (F40) FORTRAN and new FORTRAN-10 procedures can be handled.
During execution of the FORTRAN procedure, YFOXCB in the static area has
the XCB value corresponding to the procedure block.
More information is available in the PHFO module listing.
.page
III.1.10##SIMSET and SIMULATION
.br;##########---------------------
.s;The standard classes SIMSET and SIMULATION are implemented in the SS and
the SU modules, respectively, where prototypes, symbol tables and RTS
procedures implementing the main classes and their attributes, are found.
Names for quantities referred by the prototypes are generated in SIMRPA
in the same way as for the FILE subclasses, see III.1.7.1.
The classes involved are represented by the symbols (fixup numbers used by
the compiler): SSST (SIMSET), SSLG (LINKAGE), SSHD (HEAD), SSLK (LINK),
SUSI (SIMULATION), SUPS (PROCESS), SUMA (Main Program of SIMULATION).
D%SUSI is thus the macro for implementing the SIMULATION prototype,
SUSI%S is the statement coding (SUSI.ZCPSTA) of SIMULATION, etc.
The present RTS design places SS and SU in the low segment to be loaded
with the compiled SIMULA program. Since some information in the
class prototypes is dependent upon the level where the SIMULA program
uses SIMSET or SIMULATION as a prefix, the SIMULA compiler must
supply this level as global symbols to be used for relocation by
LINK-10 or the LOADER. Because both the effective block level
and its negation are involved, two symbols, .SIMLV and .SIMVL, are defined.
_.SIMVL defines ZPREBL of the SIMSET and SIMULATION prototype, i e the
offset from the block address where the display element pointing to the
class object itself can be found at run time.
_.SIMLV is the complement of .SIMVL, i e a positive quantity, which is used to define the
display vector lengths (ZPCDLE) of the classes involved.
_.SIMVL is also used as ZCPSBL (static environment block level) for the
prototypes of SIMSET and SIMULATION attributes.
.p 0,2,10;III.1.10.1##SIMSET implementation
.s;SIMSET is defined in CBL 14.1. The implementation is a straightforward
translation of this definition. FIRST, LAST and PREV (recommended extension) are
implemented inline by the compiler.
The SIMSET prototype is SSST (entry .SSST). No special actions are performed
for declarations (SSST%D) or initial actions (SSST%S). Coding to set
up a block prefixed by SIMSET is no different from any prefixed block
setup by CPSP.
.p 0,2,10;III.1.10.1.1##Classes local to SIMSET
.s;SIMSET has the class attributes LINKAGE, LINKAGE class HEAD and LINKAGE class
LINK.
LINKAGE (SSLG) objects have the two links ZLGPRE and ZLGSUC implementing
the inaccessible variables SUC and PRED defined in CBL 14.1.2.
ZLGSUC and ZLGPRE share a 36-bit word.
No additional data is needed for HEAD and LINK.
.p 0,2,10;III.1.10.1.2##Procedure attributes of LINKAGE, HEAD and LINK
.s;The procedure attributes are implemented in a straightforward manner following
CBL definitions. Parameters to pure procedures are passed in XWAC1, XWAC2, etc,
and parameters to functions are passed in ac's starting with XWACn,
where XTAC has the address of XWACn.
.p 0,2,10;III.1.10.2##SIMULATION
.s;The standard class SIMULATION is defined in CBL 14.2.
The SU module contains prototypes and actions of SIMULATION, PROCESS
and MAIN PROGRAM, procedure attributes of SIMULATION and
procedures to implement the activation statement.
The sequencing set is implemented as a binary tree of EVENT NOTICES (ZEV).
Event notices are referred by process objects (and the event notice refers
back to the process) which are scheduled. For storage management reasons,
event notices are contained in event notice records (ZER) and chained to
free lists when not in use. The event notice records are chained to
the SIMULATION block and to each other. If several SIMULATION blocks
exist, they are chained to each other for garbage collection reasons.
The SQS tree is rooted at the last eventnotice (largest scheduled time),
which is located via ZSULT of the SIMULATION block.
The current event notice (smallest scheduled time) is found via ZSUFT.
The three links ZEVZBL (upward link), ZEVZLL (left) and ZEVZRL (right)
are used to define the tree in such a way that the EVTIME (ZEVTIM) values
of the left hand subtree are always smaller than those of the
right hand subtree and thus the path from last to first event is via left links.
ZEVZBL is used for backtracking when the left or right link does not exist.
Depending on the order of scheduling statements, several possible trees
exist for a certain sequence of events. Example:
.p 0,2,16;.nf;
First Last
(ZSUFT) (ZSULT) (dummy)
------ ------ ------ ------ ------ ------
[ 1 ]<--[ 2 ]<--[ 3 ]<--[ 8 ]<--[ 10 ]<--[ZZZZ]
------ ------ ------ ------ ------ ------
! !
V V
------ ------ ------- ------
[ 4 ]<--[ 6 ]<--[ 7 ] [ 9 ]
------ ------ ------- ------
!
V
------
[ 5 ]
------
.s;.fill
where left arrows denote left links and down arrows denote right links.
ZSUFT (current) refers to the ZEV labeled 1, and the root is the last event
(10).
ZZZZ stands for the largest possible real value [3777777777777 octal] and
this node is placed in the tree to facilitate ranking of times.
The tree is manipulated at the request of scheduling statements:
ACTIVATE/REACTIVATE, HOLD, PASSIVATE, CANCEL, WAIT.
Eventnotices are allocated by SANE, which allocates a new eventnotice record
(ZER) when no more free eventnotices are found in the current record.
Special measures have to be taken to preserve event notice addresses
over a possible garbage collection. Since the garbage collector
can only relocate addresses that point to the start of recognizable
records, i e with a ZDN part, eventnotice addresses in global
locations have to be split into two parts, ZER address + offset, before
allocating a new ZER record. The global locations that can be used are
YSUPCP and YSUSCP. See SU and SANE coding.
The address of the (current) SIMULATION block is found via
the global location YSULEV in the static area, which is set up by the
initial actions of SIMULATION to contain the instruction
MOVE XSAC,.SIMVL(XCB). The effective block level of SIMULATION is
thus found in the right half of this word.
The procedure ACTIVATE described in CBL 14.2 is implemented as SUAC.
CURRENT, IDLE, MAIN, TERMINATED and TIME are implemented inline.
The rest is straightforward translation of the CBL definition, where the SQS
is implemented as described above without referring to the CBL definition
of class EVENT NOTICE.
To simplify SQS handling, a dummy ZEV with ZEVTIM=max real value is inserted
at the start. ZEVTIM is represented in single precision floating point format.
SUAC is called with an activation mask in X0, process reference in XWAC1,
second parameter (process or time) in XWAC2. See SU module for details.
.pg;III.1.11##Standard procedures and functions
.br;##########---------------------------------
.s;The following standard procedures and functions are available in DEC-10 SIMULA:
.p 0,2,10;.nofill
SIMULA name Fixup (1) Access (2) result param. call
type types (3)
.s;ABS inline G R/L R/L
ACCUM SUAM SU - R,R,R,R (5)
ARCCOS MAAC G R R F
ARCSIN MAAS G R R F
ARCTAN MAAT G R R F
MAATD G L L F
BLANKS TXBL G T I A2
BREAKOUTIMAGE IOBO IOOU P
CALL CPCA G - O A1
CANCEL SUCA SU - SUPS P
CARDINAL SSCA SSHD I A1
CHAR inline G C I
CLEAR SSCL SSHD P
CLOSE IOCL IOFI P
COPY TXCY G T T A2
COS MACO G R R F
MACOD G L L F
COSH MACH G R R F
CURRENT inline SU SUPS
DETACH CPDT G P
DIGIT inline G B C
DISCRETE RDDI G I A,I R1
DRAW RDDR G B R,I R1
EMPTY SSEY SSHD B A1
EJECT IOEJ IOPF P
ENDFILE inline IOIN,IODF B
ENTIER inline G I L/R
EVTIME SUEV SUPS R A1
ERLANG RDER G R R,R,I R1
EXP MAEX G R R F
MAEXD G L L F
FIRST inline SSHD SSLK
FOLLOW SSFW SSLK - SSLG P
GETCHAR IOGC TEXT C A1
GETFRAC IOGF TEXT I A1
GETINT IOGI TEXT I A1
GETREAL IOGR TEXT L A1
HISTD RDHI G I A,I A1
HISTO RDHO G - A,A,R,R P
HOLD SUHO SUPS - R P
IDLE inline SUPS B
INCHAR IOII IOIN,IODF C I1
INFRAC IOIF IOIN,IODF I I1
INIMAGE IOIG IOIN,IODF I1
ININT IOIG IOIN,IODF I I1
INREAL IOIR IOIN,IODF L I1
INTEXT IOIT IOIN,IODF T I I2
INTO SSIT SSLK SSHD P
Integer to MACL G L I F
long real
LAST inline SSHD SSLK
LASTITEM IOLI IOIN,IODF B I1
LETTER inline G B I
LINEAR RDLI G R A,A,I R1
LINE inline IOPF I
LINESPERPAGE IOLP IOPF - I P
LOCATE IOLT IODF - I P
LOCATION inline IODF I
Long real to MACI G I L F
integer
LOWTEN TXLT G - C P
MAIN TXMN TEXT T T1
MOD inline G I I,I
MORE inline TEXT B
inline IOFI B
NEGEXP RDNE G R R,I R1
NEXTEV SUNE SUPS SUPS A1
NORMAL RDNO G R R,R,I R1
OPEN IOOP IOFI - T P
OUT SSOU SSLK P
OUTCHAR IOOC IOOU,IODF - C P1
OUTFIX IOOX IOOU,IODF - L,I,I P1
OUTFRAC IOOF IOOU,IODF - I,I,I P1
OUTIMAGE IOOG IOOU,IODF - P
OUTREAL IOOR IOOU,IODF - L,I,I P1
OUTTEXT IOOT IOOU,IODF - T P1
PASSIVATE SUPA SUPS - P
POISSON RDPO G I R,I R1
POS inline TEXT,IOFI I
PRECEDE SSPC SSLK - SSLG P
PRED SSPD SSLG SSLK A1
PREV inline SSLG SSLG
PUTCHAR TXPC TEXT - C T1
PUTFIX TXPX TEXT - L,I T1
PUTINT TXPI TEXT - I T1
PUTFRAC TXPF TEXT - I,I T1
PUTREAL TXPR TEXT - L,I T1
RANDINT RDRA G I I,I,I R1
RANK inline G I C
RESUME CPRS G - O P
SETPOS TXSE TEXT - I P
TXSE IOFI (6) - I P
SIGN inline G I R/L
SIN MASI G R R F
MASID G L L F
SINH MASH G R R F
SQRT MASQ G R R F
MASQD G L L F
STRIP TXST TEXT T T A1
SUB TXSU TEXT T I,I A1
SUC SSSC SSLG SSLK A1
TAN MATA G R R F
TANH MATH G R R F
TERMINATED inline SUPS B
Text assignment TXVA G - T,T T4
Text relation TXRE G B T,T T4
TIME inline SU R
UNIFORM RDUN G R R,R,I R1
WAIT SUWA SUPS - SSHD P
** (power) MARR G R R,R F
MALL G L L,L F
MARI G R R,I F
MALI G L L,I F
.fill
.p 0,2,10;Notes:
.s;Type codes used: B=Boolean, C=character, A=real array, L=long real,
I=integer, O=ref(any class), R=real.
SUPS etc - see note (2).
.s;(1) Fixup is the identifying code given to the symbol by the initial
expansion of the RTSYMBOLS macro in SIMMCR. Values < 400000 octal are used as
fixup numbers for external lowseg references. Values > 400000 are used
as absolute addresses in the high segment, referring to the transfer vector in
SIMRn0 (SIMRn1 and SIMRn2).
.s;(2) If the access code is G, the procedure is available in any SIMULA program
and not an attribute. Procedures marked SU are available only in SIMULATION
blocks. TEXT signifies a text attribute, and the other four-letter codes
signify attributes to the corresponding classes:
.br;FILE (IOFI), INFILE (IOIN), OUTFILE (IOOU), DIRECTFILE (IODF), PRINTFILE (IOPF),
LINKAGE (SSLG), LINK (SSLK), HEAD (SSHD), PROCESS (SUPS).
.s;(3) Calling sequences:
.s;In all cases but F (FORTRAN sequence), parameters are in successive ac's
starting with the current top of stack ac, called Xtop.
For pure procedures and for most function calls, Xtop=XWAC1. Most
procedures are optimized for this case and are coded to place the parameters
in standard ac's before processing and restore ac's on exit.
In the case of text and class attributes, the first actual parameter
(in Xtop and Xtop+1) is the class or text reference. The result is placed
in Xtop (+Xtop+1) or where Xtop points.
.s;A1 is used for most functions which are attributes of a class or of TEXT:
.nf;.lm 16
[load address of top ac to XTAC]
[object reference in Xtop or text reference in Xtop and Xtop+1]
PUSHJ###XPDP,f
.f;.lm 8
where f is the function fixup code.
.s;A2 is used for a global function when garbage collection may occur:
.nf;.lm 16
[Any arguments to Xtop, Xtop+1 etc]
PUSHJ XPDP,f
XWD n,admap
.f;.lm 8
where n is Xtop-XWAC1 and admap points to a bit map showing which ac's
XWAC1 etc contain pointers. Admap is zero if n is zero.
.s;F is the calling sequence used for arithmetic functions taken from the
FORTRAN library and for TAN (copied from ALGLIB):
.nf;.lm 16
[Put the first parameter in .YFARG]
[any second parameter in .YFAR2]
MOVEI XFP,.YFADR
PUSHJ XPDP,f
(D)MOVE Xtop,X0
.f;.lm 8
where XFP=X16, and .YFADR, .YFADR+1 have the addresses of .YFARG, .YFAR2.
.s;I1 is used for certain I/O functions:
.nf;.lm 16
[File ref in Xtop]
[parameters in Xtop+2, Xtop+3, ...]
MOVEI XTAC,Xtop
PUSHJ XPDP,f
.f;.lm 8
Note that Xtop+1 is not used. Xtop.image.pos is affected.
.s;I2 is similar to A2, with a file reference in Xtop, first parameter in Xtop+1.
.s;P is the sequence for a pure procedure statement:
.nf;.lm 16
[object reference or first parameter in XWAC1]
[Rest of parameters in XWAC2, ...]
PUSHJ XPDP,p
.f;.lm 8
where p is the procedure address.
.s;P1 is used for output editing procedures:
.nf;.lm 16
[File ref in XWAC1]
[parameters in XWAC3, XWAC4, ...]
PUSHJ XPDP,p
.f;.lm 8
.s;T1 corresponds to P1 for text editing, with XWAC1 a pointer to
a text descriptor (ZTV instance).
.s;R1 is used for random drawing functions. As A1 except that the last integer
parameter (random stream number) must be an integer variable (possibly value
parameter).
.s;.s;
(5) The interpretation of "NAME" for the 3 first parameters is
"address of cell".
.s;(6) <file ref>.setpos is implemented as <file ref>.image.setpos.
.pg;.nm ch 2
.ch SIMULA FOR DEC SYSTEM 10#############TD, RTS
.nm 1
III.2##RTS DECOMPOSITION
.s;The RTS is coded as several source modules, each containing one or more
RTS subroutines or system class prototypes. The source modules, when compiled
by MACRO-10 together with the required UNIVERSAL and other parameter files,
yield REL files which are combined into:
.s;1) SIMLIB.REL, created by FUDGE2 (or MAKLIB) from those REL files
compiled with a zero origin (relocatable) and thus meant for the low
segment, plus some chosen routines from SYS:FORLIB.
.s;2) SIMRTS, the high segment, which may exist as SIMRn0.EXE
(n = SIMULA version number, e.g_. 4)
in a one high segment version, or more commonly as
.br;2a) SIMRn1.EXE, containing the most frequently used high segment routines. Created by LOAD - SSAVE.
.br;2b) SIMRn2.EXE, which contains routines for program initialization, file
creation and initialization. Created by LOAD - SSAVE.
SIMRn1 and SIMRn2 are not in core simultaneously. The transfer from one to the
other is handled by the transfer vector and the OCSW routine (see III.1.6.2).
Some routines are included both in SIMRn1 and SIMRn2.
.s;3) SIMDDn.ABS, containing self-relocating code for the debugging sub-system.
Created by executing SUTABS.REL loaded with SIMDDT.REL.
.s;4) SIMHGH.REL, containing the same REL files as
used for creating SIMRn0.EXE (those marked 0 or 1,2 in III.2.2).
To be used for debugging or for loading a SIMULA program (partly)
in the high segment.
.pg;III.2.1##SIMLIB components
.br;#########-----------------
.s;.nofill
Module Global symbols
.s;OCSP .OCSP
PHPT .PHPT
MA .MACI .MACL
RDDI .RDDI
RDDR .RDDR
RDER .RDER
RDHI .RDHI
RDHO .RDHO
RDLI .RDLI
RDNE .RDNE
RDPO .RDPO
RDNO .RDNO
RDRA .RDRA
RDUN .RDUN
TAN TAN.
SU .SUSI
SS .SSST
SANE .SANE
PHFO .PHFO
[The following routines are taken from SYS:FORLIB]
ACOS. ACOS.
ASIN. ASIN.
ATAN. ATAN.
SQRT. SQRT.
SINH. SINH.
COSH. COSH.
TANH. TANH.
EXP1 EXP1. EXP1.0 EXP1.2 EXP1.4 EXP1.6
EXP3 EXP3. EXP3.. EXP3.0 EXP3.2 EXP3.4 EXP3.6
EXP2 EXP2. EXP2.. EXP2.0 EXP2.2 EXP2.4 EXP2.6
EXP. EXP.
ALOG. ALG10. ALOG.
SIN. COS. COSD. SIN. SIND.
DEXP.2 DEXP.2 DEXP2.
DEXP.3 DEXP.3 DEXP3.
DEXP. DEXP.
DLOG. DLG10. DLOG.
DSQRT. DSQRT.
DSIN. DCOS. DSIN.
DATAN DATAN.
FORXIT EXIT
CFRXIT CEXIT.
EXIDOT EXIT.
.fill
.s;OCSP is loaded with all SIMULA main programs, the others on demand by library
search. OCSP loads the initial high segment.
.pg;III.2.2##SIMRTS components
.br;#########-----------------
.s;The following is contained in SIMRTS:
.s;.nofill
Module Segment Global symbols
.s;
SIMRTS 0 .IODF .IOFI .IOIN .IOOU
.IOPF .PDERR
SIMRT1 1 .IODF .IOFI .IOIN .IOOU
.IOPF .OCSW .PDERR
SIMRT2 2 .IODF .IOFI .IOIN .IOOU
.IOPF .OCSW .PDERR
CS 1 .CSCA .CSEN .CSEP .CSER
.CSES .CSEU .CSGO .CSNA
.CSQU .CSRA .CSSA .CSSA.
.CSSB .CSSC .CSSN .CSSW
.CSSW0
CSSADM 2 .CSRA .CSSA.
CP 1 .CPCA .CPCD .CPCI .CPDT
.CPE0 .CPNE .CPPD .CPRS
.CPSP
IO 1 IODF%D IODF%I IODF%M IODF%S
IODF%Y IOFI%D IOFI%I IOFI%M
IOFI%S IOFI%Y IOIN%D IOIN%I
IOIN%M IOIN%S IOIN%Y IOOU%D
IOOU%I IOOU%M IOOU%S IOOU%Y
IOPF%D IOPF%I IOPF%M IOPF%S
IOPF%Y .IOBO .IOCS .IOCL
.IOCLA .IOCS .IOEJ .IOFD
.IOIG .IOLI .IOLN .IOLP
.IOLT .IONB .IOOG .IOOP
.IORB .IOSP
IOED 1 IOTXR .IOIC .IOIF .IOII
.IOIR .IOIT .IOOC .IOOF
.IOOI .IOOR .IOOT .IOOX
IONF 2 .IOCF .IOCOM .IOENT .IOLOK
.OCINE .OCINK
PH 1 .PHCV .PHFA .PHFM .PHFS
.PHFT .PHFV
SA 1,2 .SAAB .SAAR .SACL .SADB
.SADE .SAGC .SAGI .SAIN
.SANP
TX 1 .TXCY .TXDA .TXGC .TXGF
.TXGI .TXGR .TXLT .TXMN
.TXPC .TXPF .TXPI .TXPR
.TXPX .TXRE .TXSE
TXBL 1,2 .TXBL .TXST .TXSU .TXVA
OCIN 2 YOCSWL .OCIN .OCIN4 .OCIN5
.OCIN6 .OCIN7 .OCINB .OCINF
.OCINI .OCINJ .OCINY .OCINZ
OCIO 1,2 .IOERF .IOFER .IOTYS .OC8T
.OCDT .OCIN1 .OCIN2 .OCIN8
.OCIN9 .OCINC .OCIND .OCING
.OCINH .OCLA .OCOO
OCEP 1 .FORER .FOREX .OCEI .OCEP
.OCLD .OCTR .OCUU
REL:HELPER.REL 2 .HELPR
.fill
.s;Modules marked 0 are only used in the one high segment version SIMRn0.EXE.
Those marked 1 are used in SIMRn1.EXE, those marked 2 in SIMRn2.EXE.
All modules except CSSADM.REL, SIMRT1.REL and SIMRT2.REL are used in SIMRn0.EXE.
.pg;.nm ch 3
.ch SIMULA FOR DEC SYSTEM 10#############TD, RTS
.nm 1
III.3##LOW SEGMENT ORGANIZATION
.s;Normally (1), the low segment of an executing SIMULA program contains:
.fn 5;.s;.c;--------------
.br;(1) Part or all of the program can also be loaded in the high segment
if a shared SIMULA program is desired or if the program is very big and
more space in the low segment is needed for data. See SIMHGH.HLP.
Also useful for debugging the RTS.
!
.s;1) The compiled code of the SIMULA main program.
.s;2) Code for any external classes or procedures, possibly
compiled by MACRO-10 (or an equivalent processor, e g BLISS) or FORTRAN.
.s;3) RTS routines loaded from SIMLIB by library search. OCSP is loaded with
all SIMULA programs, the other routines are loaded as needed.
.s;4) If SIMDDT is loaded directly at the start of execution (DEBUG or
LOAD - REENTER), it occupies the space directly following the last routine
loaded by LINK-10 or LOADER.
.s;5) The so-called STATIC area is allocated next. See SIMRPA. The static area
is addressed by loading the right half of .JBOPS to a register, usually called
XLOW (default XLOW=X16) and using the symbols declared via the STATIC macro
in SIMRPA as offsets.
The RTS stack based on XPDP is part of the static area.
.s;6) The dynamic low segment area starts with
.br;6a) an area for buffers and buffer headers,
.br;6b) the IOSPEC table,
.br;6c) more buffers if needed.
.br;See IO.MAC for table layouts and record definitions.
.s;7) The storage pool for dynamically allocated objects is at the end of the
low segment. SIMDDT and its work areas may be allocated in the pool if
not loaded at program start.
.pg;III.3.1##Low segment layout
.br;#########------------------
.s;.nofill
------------------------- <--- 0
! JOBDAT area !
!-----------------------! <--- 140 (octal)
! main !
! program !
!-----------------------!
! external procedures !
! and classes !
!-----------------------! <--- OCSP
! RTS !
! routines !
!-----------------------! <--- YOCDDT if non-zero
_. (SIMDDT if loaded) .
:-----------------------: <--- .JBSA, .JBOPS
! Static area !
!-----------------------! <--- YOCBST
! Initial buffer !
! area !
!-----------------------! <--- YIOSPC
! IOSPEC table !
!-----------------------!
! Buffer area !
! continued !
!-----------------------! <--- YSABOT
! Storage pool !
! for SIMULA !
! objects !
! - - - - - - - - - - - ! <--- YSATOP
! Unused part of !
! pool !
!- - - - - - -! <--- YSALIM
! Space for a ZAC obj. !
------------------------- <--- .JBREL
.fill
.pg;.nm ch 4
.ch SIMULA FOR DEC SYSTEM 10#############TD, RTS
.nm 1
III.4##RTS INTERFACE
.s;The RTS interfaces with a) the user,
b) the monitor and file system, c) the compiled program.
Within the RTS, there are interfaces d) between low and high
segment, e) between SIMDDT and the rest of the RTS.
.p 0,2,10;III.4.1##RTS interface with the user
.br;#########---------------------------
.s;The user communicates with the RTS by the following means:
.s;- The user TTY, which is used by the RTS for error messages and other
diagnostics, for the file initialization dialogue if specified
(III.1.7), to receive SIMDDT commands, etc. The TTY is used both as
SYSIN and as SYSOUT if no other specification has been given.
.s;- A specification file may be read if the /R switch is given at
compile-time.
See III.1.7.
.p 0,2,5;III.4.2##RTS interface with monitor and file system
.br;#########------------------------------------------
.s;
This is described mainly in III.1.6 and III.1.7.
.p 0,2,10;III.4.3##RTS interface with compiled program and internally
.br;#########--------------------------------------------------
.s;SIMRTS, the high segment of RTS, interfaces with the compiled program
via the transfer vector and the static area.
The low segment of the RTS interfaces with compiled program via global symbols
and the static area.
.p 0,2,8;III.4.3.1##Low segment - high segment interface
.s;Since the high segment (SIMRTS) is created independently of any
low segment that is present at run time, communication between the
segments must go through fixed addresses or via a base register
and fixed offsets. Both methods are used, as explained below.
.p 0,2,10;III.4.3.1.1##References to high segment from low segment
.s;High segment routines are called via a transfer vector, consisting of
JRST and PUSHJ instructions. The transfer vector is created
by expansion of the RTSYMBOLS macro, with a suitable
definition of the macro which is called for each entry. No global
variables are placed in the high segment.
.p 0,2,10;III.4.3.2##References from the high segment to the low segment
.s;The (static) area for global variables and tables is placed at the initial .JBFF
location. .OCSP saves the address of the global area in .JBOPS, right half.
When referring from either the high or the low segment to those global
locations, a base register must be used, which is loaded by a HRRZ instruction from .JBOPS when needed. The symbolic name of that register should always be XLOW. The actual value of XLOW may change throughout the run time
system, but usually XLOW=XFP is used.
The global variables are thus always indexed by XLOW.
.s;Some low segment data must be loaded together with the SIMULA program.
These are:
.s;* Pseudo or extended accumulators which are
global locations defined at load time, with the names .YXA1, .YXA2, etc.
The compiler refers to those locations by consecutive fixup numbers,
which are translated to external references. In order to have access to
those locations also from the high segment, the address of .YXA1
must be placed in the static area based on XLOW. YXACAD(XLOW) contains that address.
.s;* Address and value list for communication with FORTRAN library
subroutines, e g SIN, COS, and the exponentiation routines. The address
list consists of two words at address .YFADR. Those words have the
addresses of the two doublewords .YFARG and .YFAR2.
When calling, e g, the SQRT function, the argument is placed in .YFARG
and XFP is made to point to .YFADR, before executing the PUSHJ.
As an example, the following calling sequence is generated for the
statement y:=sqrt(x):
.nf;.p 0,1,7
MOVE XWAC1,x
MOVEM XWAC1,.YFARG
MOVEI XFP,.YFADR
PUSHJ XPDP,SQRT. (fixup MASQ in RTSYMBOLS)
MOVE XWAC1,0
MOVEM XWAC1,y
.fill
.p 0,2,10;III.4.3.3##References from the RTS to the compiled program
.s;SIMDDT uses the main line number table address to find line number
and symbol tables for the main program and any external procedures or classes.
The RTS stack, starting at YOBJRT of the static area, is used for
referring to compiled code in several RTS routines and in SIMDDT.
_.JBSA points either to .MAIN or .OCRE0 (DEBUG command).
The data area for the outer program block and its display vector are placed
in compiled code. Program objects (subblocks, procedures, classes, prefixed
blocks) refer to compiled code via the prototype pointer.
Other blocks may also contain references to compiled code.
The SIMULATION or SIMSET block level is defined by global symbols (see III.10).
.p 0,2,10;III.4.3.4##Access to SYSIN, SYSOUT objects, to SIMULATION block
.s;The SYSIN file object address is found at offset YSYSIN in the static area.
Code is compiled to load an ac from .JBOPS before accessing SYSIN.
Access to SYSOUT is via YSYSOUT.
In a SIMULATION program, YSULEV contains information (MOVE XSAC,offset of
simulation block address in current display) to find the SIMULATION block.
Since these three offsets are used by compiled code, they must not be changed
unless absolutely necessary (old SIMULA rel files will be useless).
.pg;.nm ch 5
.ch SIMULA FOR DEC SYSTEM 10#############TD, RTS
.nm 1
III.5##COMMON MACRO PARAMETER FILES
.s;MACRO-10 modules in the RTS refer to the UNIVERSAL files SIMMAC.UNV,
SIMMCR.UNV and SIMRPA.UNV, created from the corresponding _.MAC files.
SIMMAC is described elsewhere (I.6).
SIMMCR contains information used at code generation, chiefly record definitions
for static and dynamic records used at run time.
SIMRPA contains additional definitions used mostly at run time but also
for code generation in some cases.
.p 0,2,10;III.5.1##UNIVERSAL SIMMCR
.br;#########----------------
.s;SIMMCR contains the following:
.br;- Definitions of register names XIAC, XFP, XCB, XSAC, XTAC, XRAC, XRAC1.
.br;- TYPZDN macro, used to define the constants for run-time dynamic record
discrimination (ZDNTYP field) and also in the garbage collector to define
a jump table for record processing.
The last tag value used, QZDNTM, is defined.
.br;- RTS error codes used at compile time.
.br;- Parameter descriptor codes used in ZFL and ZAP records.
.br;- Record definitions (DR, DF) for dynamic records ZDN, ZBI, ZBP, ZCL, ZPB, ZTE,
ZTT, ZAR, ZAC, ZER, ZDR, ZYS, ZXB, ZTV, ZDV, ZDS, ZLD, ZDL, ZDA, ZRV, ZDP,
ZFL, ZSU, ZEV, ZLG, ZPS,
static records ZPR, ZPC, ZCP, ZMP, ZSL, ZSR, ZTD, ZPD, ZFP, ZFR, ZAP,
ZLN, ZSD, ZSM, ZTH and miscellaneous definitions: ZTS (thunk save area),
YFOFAD, YFOAAD, YFOPAD.
.br;- The RTSYMBOLS macro, which gives values to most symbols used to access
RTS routines and system class prototypes.
The entries in the RTSYMBOLS macro are calls of an X macro, used for
procedure entries, and a Y macro, used for prototypes.
The X macro is invoked:
.br;########X(ROUT,SEG,SIM,N)
.br;ROUT is the mnemonic used for the entry.
SEG is L (low segment), 1 (SIMRn1), 2 (SIMRn2) or 3 (both SIMRn1 and SIMRn2).
An empty second parameter stands for 1.
SIM is the letter S if the symbol has a SIMULA name (exists in initial
symbol table), empty otherwise.
N, if present, gives the global name to be used for the entry
point to the routine, otherwise the global name is formed by prefixing ROUT with a dot.
.br;The Y macro is invoked:
.br;########Y(PROT,SEG,SIM,F)
.br;PROT corresponds to ROUT. If SEG is 1,2 or 3, the prototype will be expanded
in the transfer vector directly to be accessible to compiled code via PROT.
The initial expansion reserves space for the prototype to get the correct
values of the following entries.
F, if present, shows the size of the variable part of the prototype (after
the fixed ZPC part, i e parameter descriptors for formal parameters).
RTSYMBOLS, when expanded with the initial definitions of X and Y, defines
fixed high segment addresses if SEG=1,2 or 3, consecutive fixup numbers
starting with 1 if SEG=L.
The high segment addresses are used directly in compiled code as
entry addresses to RTS procedures and prototypes. The fixup numbers are
coupled with an external symbol table defined in PASS 3 of the compiler
by another expansion of RTSYMBOLS (with X and Y redefined).
In the SIMRTS module used to produce SIMRTS.REL, SIMRT1.REL or SIMRT2.REL,
another expansion of RTSYMBOLS produces the transfer vector used at
run time.
RTSYMBOLS also defines fixup numbers for pseudo ac's and the FORTRAN
subroutine interface cell - .YFADR, .YFARG, .YFAR2.
A readable representation of RTSYMBOLS is produced by compilation
(with listing) of the following MACRO-10 code:
.s;.nofill
TITLE RTSYMBOL VALUES
SEARCH SIMMAC,SIMMCR
IF2,<END>
DEFINE XX(NA,VAL,S1,EN)<
PRINTX NA VAL S1 EN
>
DEFINE X(ROUT,S,DUM,N)<
IFNB <N>,<XX(ROUT,_\ROUT,S,N)>
IFB <N>,<XX(ROUT,_\ROUT,S,.'ROUT)>
>
DEFINE Y(PROT,SEG,S,F)<
XX(PROT,_\PROT,SEG,.'PROT)
>
PRINTX SYMBOL VALUE SEG ENTRY
PRINTX ------------------------------
RTSYMBOLS
END
.fill
.pg;III.5.2##UNIVERSAL SIMRPA
.br;#########----------------
.s;SIMRPA contains definitions needed mostly in the RTS and in some
compiler modules.
.p 0,2,10;III.5.2.1##SIMRPA macros
.s;The following macros are defined:
.s;LOWADR(X) - defines XLOW=X and loads XLOW from .JBOPS.
.s;SETLOW(X) - defines XLOW=X, does not change X.
.s;IFNONE(X) - skips if X==NONE.
.s;TRIMSTACK - removes top of XPDP stack.
.s;ERRMAC(A) - defines the error message macro to be used in component A (A'ERR).
.s;RIGHTHALF(A) - compile time message if A is not a right half field,
where A is a field designator.
.s;SAVEALLACS - All ac's are saved at offset YUUOAC of the static area.
.s;CALLOW, CDEFER, CENABLE, CFORBID - control _^C-REENTER SIMDDT interface
(see III.1.6.2.3.3).
.s;TEXT(T), RTEXT(T), RRTEXT(T) - used for garbage collector debugging.
.s;ERRCODE(N0,A) - Defines the initial error number used in module A, Q'A'ENO=N0.
.s;RTSMODULE(M,M1,SEG) - Defines TITLE, ERRMAC(M), high or low segment
relocation.
.s;PROCINIT(A) - Defines declarations for IO, OCIO, OCIN modules when expanded.
A is module name.
.s;RDINIT(e) - Defines macros and initial info for submodule RD'e
of RD.MAC (submodules are separated by PRGEND).
.s;DCLASS(NAM,...) - Defines D%'NAM (prototype macro) for the standard
class with index code NAM, e g IOFI (FILE).
.s;DSZCML - Defines symbol table record (ZSM) for a standard class.
.s;DZSD - Defines symbol table entry for local quantity.
.s;SYSCLASS - Used to define codes QCLPB, QSUSI, QSSLG, QSUPS, QIOFI for the
ZCPGCI field in the prototype. Used by garbage collector to indicate
special relocation. SYSCLASS is used there to define a jump table.
.p 0,2,10;III.5.2.2##Constants defined in SIMRPA
.s;Alternative ac names are defined.
Compile time constants like QIOLP (default LINESPERPAGE value).
Entry point offsets for SIMDDT.
Constants referring to monitor tables, FORTRAN error codes, interrupt
conditions.
.s;The offsets in the STATIC area are defined via the STATIC macro.
These offsets begin with the letter Y.
.p 0,2,6;III.5.2.3##Record definitions in SIMRPA
.s;The following records are defined:
.br;ZFS, ZFI, ZIF, ZOF, ZPF, ZDF, ZBH, ZSW, ZBU, ZYS as SFD block, ZXB, ZFD.
.p 0,2,5;III.5.2.4##Switch definitions in SIMRPA
.s;Switches used in OC, IO and SA modules are defined.
.pg;.nm ch 6
.ch SIMULA FOR DEC SYSTEM 10#############TD, RTS
.nm 1
III.6##TESTING
.s;Special debugging tools have been developed for the I/O subsystem (see I.10,
test standard), and the garbage collector (SAGC, III.6.2).
For the rest of the RTS, the standard debugging tool, DDT has been used.
SIMDDT can also be used (in conjunction with DDT).
At present (version 4 of SIMULA) the debugging tools for SAGC cannot
be used, however.
.p 0,2,10;III.6.1##Debugging with DDT
.br;#########------------------
.s;III.6.1.1##Debugging using rel files for all RTS components
.s;DDT debugging is simplified if all REL files both for low and high segment
are available. In this case the required files are loaded directly and possibly
saved as an EXE file. SIMLIB may be used in library search mode.
Note that local symbols are not available for routines in SIMLIB.
If all required rel files are loaded directly with DDT via a DEBUG command,
breakpoints may be set anywhere before starting the program.
The following command sequence may be used:
.nofill
.DEBUG SIMPRO.SIM,/REL REL:SIMHGH.REL,REL:HELPER.REL
[set breakpoints]
_^[G (or .OCRE0_^[G if SIMDDT start is wanted)
[execution]
.fill
where _^[ stands for <altmode> and SIMHGH.REL contains the modules
SIMRTS, CS, CP, IO, IOED, OCIN, OCIO, PH, SA, TX, OCEP.
The REL files from SIMHGH.REL may also be loaded separately.
SIMRTS.REL (the one-segment version)
must always be the first module in the high segment.
If possible, REL files compiled with QDEBUG=1 and QSYS=0 should be used
(see implementation guide).
For complete garbage collector debugging, SADEB.REL must be included.
See III.6.2.
If execution is made to start in SIMDDT, DDT may be entered by
typing _^C when a command is expected, then issuing the DDT command.
Control is returned to SIMDDT by executing "JRST 2,@.JBOPC_^[X" and then
typing a SIMDDT command.
Note that SIMDDT may be used to force a garbage collection via the
VARIABLES command. This is useful when a garbage collection error is expected.
See III.6.2 for ways to dump SIMULA objects and to trace garbage collector actions.
.p 0,2,10;III.6.1.2##Debugging the SYS version of the RTS
.s;Debugging the SYS version of the RTS high segment is rather difficult.
If possible, the rel files should be used as described in III.6.1.1.
With DDT, breakpoints can be set in the low segments without difficulty, but
breakpoints cannot be set in the high segment at all if the SHR files from
SYS are used at user level. One possible way is to GET and SAVE SIMRn1 and
SIMRn2 from SYS, creating unsharable EXE files on the user area, then load the program
with DDT (DEBUG/D SIMPRO) and SAVE it, then ASSIGN DSK SYS
(after setting the SYS switch via SETSRC).
If breakpoints must be placed in SIMRn1, set a breakpoint at .MAIN+6, then
set the required high segment breakpoints after stopping at .MAIN+6.
It is also possible to start in SIMDDT and enter DDT at a breakpoint as
shown above.
If breakpoints in SIMRn2 are needed, set a breakpoint at .OCSP+73 (may be
changed, check for JRST 400047) and start. SIMRn2 is then in core.
If you need to set breakpoints at other invocations of SIMRn1 or SIMRn2 than
the first one, patches must be made in the high segment. It is possible e g
to place "EXIT 1,", which works as _^C, in a high segment location,
noting the replaced instruction.
If the replaced instruction is is HRRZ 16,.JBOPS (LOWADR), it may be
redundant, otherwise the replaced instruction has to be executed in DDT
explicitly (_^[X command) before returning via .JBOPC as shown above.
This debugging mode should not be used unless it is necessary to reproduce
an error.
If the error can be shown to occur when all rel files are loaded,
debugging should be done as in III.6.1.
If the SYS version has only one high segment, debugging is somewhat
simplified.
.pg;III.6.2##Testing facilities in the garbage collector
.br;#########-------------------------------------------
.s;.nt WARNING
.s;The testing facilities are currently (version 4) not usable.
.en
.s;In the test version (QDEBUG=1, QSYS=0), the garbage collector (SAGC, shortened
as GC) processing can be traced. The global cell at offset YSASW in the static
area contains 5 switches to determine the amount of test output produced.
The switches are:
.s;.nofill
Switch bit no Meaning
.s;SAGCPE 0 1 if no output on nnnGCP.TMP
SWGCTE 1 1 for test output from GC
SWGCT2 2 1 for test output on TTY
SWGCT3 3 1 for test output on SYSOUT
SWGCT4 4 1 if GC runtime and low seg limit
should be logged.
.s;.fill
If SAGCPE is off, the GC parameters used in the storage allocation algorithms
are output in append mode on the file nnnGCP.TMP after each garbage collection,
where nnn is job number.
The data on this file can be analysed by the utility program
SUTGCA (see chapter V). SAGCPE is set off by default if QDEBUG=1.
Initialization is done by .SAGI, which opens the file. SAGCPE is set off if I/O errors
occur on the file.
.s;If SWGCTE is on, the test version logs GC actions. In PHASE1 the start
addresses of all chained records are output. In PHASE2 all referenced records
in the pool are logged with old address, new address, record length given.
In PHASE3 all updated pointers are given with old and new values.
Finally, in PHASE4, the old address, new address, and length for each moved
record are output.
SWGCTE is off by default.
.s;SWGCT2 and SWGCT3 specify the output destination for test output.
Default is TTY (SWGCT2 and NOT SWGCT3).
.s;If SWGCT4 is set a short log message is given on each GC execution.
.s;The default switch settings can be changed by patching the .SAGI routine.
The appropriate bits in the first instruction compiled for SETON
SWGCT2(XLOW) can be set. The bits may also be changed in YSASW(XLOW)
at a DDT breakpoint.
.p 0,2,10;The SADEB.MAC module contains a number of dump routines to be used for GC
testing as well as the output routines for logging as specified above.
.s;_.SAPD dumps the whole object pool.
.s;_SAPDRE dumps a single record in the pool.
.s;_.SQSDU dumps the sequencing set of SIMULATION (SQS).
.s;The routines are invoked by the _^[X command in SIMDDT, executing a PUSHJ 17,r
instruction, where r is the appropriate routine.
.s;When calling SAPDRE, XCUR==X6 must contain the start address of the record.
For easier handling of the dump routines, note that the DDT command
"PUSHJ 17,.SAPD<PD:", for example, defines PD as PUSHJ 17,.SAPD.
Thus PD_^[X may be used to get a pool dump.
.pg;.nm ch 7
.ch SIMULA FOR DEC SYSTEM 10#############TD, RTS
.nm 1
III.7##SAMPLE PROGRAM
.s;As an illustration of what happens at run time, the execution of the
following SIMULA program is analysed in some detail, referring to line numbers
as allocated by the compiler:
.p 0,2,10;.nofill
B1 1 BEGIN
2 CLASS pb1;;
B2 3 pb1 BEGIN
4 CLASS c1;
B3 5 BEGIN
6 CLASS c2;
B4 7 BEGIN
8 detach;
9 rc3:-NEW c3;
10 resume(rc3);
E4 11 END c2;
12 CLASS c3;
B5 13 BEGIN
14 detach;
E5 15 END c3;
16 REF(c2)rc2;
17 REF(c3)rc3;
18 CLASS pb2;;
19 rc2:-NEW c2;
20 detach;
B6 21 pb2 BEGIN
22 CLASS pb3;;
B7 23 pb3 BEGIN
24 resume(rc2);
E7 25 END pb3;
E6 26 END pb2;
E3 27 END c1;
28 REF(c1)rc1;
29 rc1:-NEW c1;
30 resume(rc1);
31 resume(rc1);
E2 32 END pb1;
E1 33 END program;
.s 2;
.fill
Tracing with SIMDDT gives the following execution order by source line number:
3, 2, 29, 19, 8, 20, 30, 21, 18, 23, 22, 24, 9, 14, 10, 15, 31,
25, 26, 27, 32.
.s;This is explained as follows:
.s;Start before line 1 at the entry point .MAIN given by .JBSA.
Call OCSP, which swaps in SIMRn2, transfers control to OCIN for initialization
(SYSIN and SYSOUT setup etc). OCIN transfers control to .OCEI via the
transfer vector in SIMRn2, swapping in SIMRn1 to replace SIMRn2 (.OCSW).
OCEI returns to compiled code.
.s;CSER is called to initialize the reduced subblock starting at line 1, using
_.SAIN to initialize data according to subblock map.
.s;At line 3, CPSP, using SADB and .SAIN, sets up the prefixed block
with its display vector and dynamic link. CSEN is called to start
declarations in the prefix (ZPCDEC of the pb1 prototype) then via
CPCD to ZPCDEC of the prefixed block, then via CPPD to line 2 (statements
of the prefix), continuing via CPCI (implicit inner) to ZCPSTA of the
block (line 29).
.s;At 29 a c1 object is generated and entered: CPNE [SADB [.SAIN] CSEN] -
ZPCDEC - CPCD - ZCPSTA (line 19).
.s;At 19, create c2 like c1, ending up at line 8.
.s;At line 8, detach (CPDT) detaches c2 and returns the reference to the caller,
continuing at line 20, where c1 is detached, returning to line 30.
.s;At line 30, resume (CPRS) of c1 leads to the reactivation point at line 21.
.s;At line 21 a prefixed block again starts via CPSP, CSEN, prefix class (line 18),
continuing at line 23, where a new prefixed block is started (lines 22, 24).
.s;At line 24 c2 is resumed at its reactivation point line 9.
C3 is resumed (line 14) and detaches to line 10.
.s;At line 10, c2 terminates via CPE0. CPE0 calls CPDT, leading to the QPS
reactivation point at line 15, where c3 terminates, detaching to line 31
of the surrounding prefixed block.
The resume (CPRS) at line 31 leads to line 25 in the innermost prefixed block
via operating chain and back down via dynamic links.
.s;The prefixed blocks terminate in turn via CPE0 and ZCPIEA links:
lines 25, 26. C1 terminates via CPE0 and CPDT, leading to
line 32, where the pb1 prefixed block terminates via CPE0 and ZCPIEA.
.s;OCEP terminates program execution, closing files via .IOCLA and .IOCL.
Returns to monitor level via EXIT 1,.