Figure 3.1: Preprocessing phases: from a user file to a source file
This paper describes high-level objects and functions that are potentially user-visible in a PIPS1 [?] interactive environment. It defines the internal software interface between a user interface and program analyses and transformations. This is clearly not a user guide but can be used as a reference guide, the best one before source code because PIPS user interfaces are very closely mapped on this document: some of their features are automatically derived from it.
Objects can be viewed and functions activated by one of PIPS existing user interfaces: tpips2 , the tty style interface which is currently recommended, pips3 [?], the old batch interface, improved by many shell scripts4 , wpips and epips, the X-Window System interfaces. The epips interface is an extension of wpips which uses Emacs to display more information in a more convenient way. Unfortunately, right now these window-based interfaces are no longer working and have been replaced by gpips. It is also possible to use PIPS through a Python API, pyps.
From a theoretical point of view, the object types and functions available in PIPS define an heterogeneous algebra with constructors (e.g. parser), extractors (e.g. prettyprinter) and operators (e.g. loop unrolling). Very few combinations of functions make sense, but many functions and object types are available. This abundance is confusing for casual and experiences users as well, and it was deemed necessary to assist them by providing default computation rules and automatic consistency management similar to make. The rule interpretor is called pipsmake6 and described in [?]. Its key concepts are the phase, which correspond to a PIPS function made user-visible, for instance, a parser, the resources, which correspond to objects used or defined by the phases, for instance, a source file or an AST (parsed code), and the virtual rules, which define the set of input resources used by a phase and the set of output resources defined by the phase. Since PIPS is an interprocedural tool, some real inpu resources are not known until execution. Some variables such as CALLERS or CALLEES can be used in virtual rules. They are expanded at execution to obtain an effective rule with the precise resources needed.
For debugging purposes and for advanced users, the precise choice and tuning of an algorithm can be made using properties. Default properties are installed with PIPS but they can be redefined, partly or entirely, by a properties.rc file located in the current directory. Properties can also be redefined from the user interfaces, for example with the command setproperty when the tpips interface is used.
As far as their static structures are concerned, most object types are described in more details in PIPS Internal Representation of Fortran and C code7 . A dynamic view is given here. In which order should functions be applied? Which object do they produce and vice-versa which function does produce such and such objects? How does PIPS cope with bottom-up and top-down interprocedurality?
Resources produced by several rules and their associated rule must be given alias names when they should be explicitly computed or activated by an interactive interfaceFI: I do not understand.. This is otherwise not relevant. The alias names are used to generate automatically header files and/or test files used by PIPS interfaces.
No more than one resource should be produced per line of rule because different files are automatically extracted from this one8 . Another caveat is that all resources whose names are suffixed with _file are considered printable or displayable, and the others are considered binary data, even though they may be ASCII strings.
This LATEX file is used by several procedures to derive some pieces of C code and ASCII files. The useful information is located in the PipsMake areas, a very simple literate programming environment... For instance alias information is used to generate automatically menus for window-based interfaces such as wpips or gpips. Object (a.k.a resource) types and functions are renamed using the alias declaration. The name space of aliases is global. All aliases must have different names. Function declarations are used to build a mapping table between function names and pointer to C functions, phases.h. Object suffixes are used to derive a header file, resources.h, with all resource names. Parts of this file are also extracted to generate on-line information for wpips and automatic completion for tpips.
The behavior of PIPS can be slightly tuned by using properties. Most properties are linked to a particular phase, for instance to prettyprint, but some are linked to PIPS infrastructure and are presented in Chapter 2.
To understand and to be able to write new rules for pipsmake, a few things need to be known.
The rule:
Properties are also declared in this file. For instance
The following variables are defined to handle interprocedurality:
These variables are used in the rule definitions and instantiated before pipsmake infers which resources are pre-requisites for a rule.
This paper also defines and describes global variables used to modify or fine tune PIPS behavior. Since global variables are useful for some purposes, but always dangerous, PIPS programmers are required to avoid them or to declare them explicitly as properties. Properties have an ASCII name and can have boolean, integer or string values.
Casual users should not use them. Some properties are modified for them by the user interface and/or the high-level functions. Some property combinations may be meaningless. More experienced users can set their values, using their names and a user interface.
Experienced users can also modify properties by inserting a file called properties.rc in their local directory. Of course, they cannot declare new properties, since they would not be recognized by the PIPS system. The local property file is read after the default property file, $PIPS_ROOT/etc/properties.rc. Some user-specified property values may be ignored because they are modified by a PIPS function before it had a chance to have any effect. Unfortunately, there is no explicit indication of usefulness for the properties in this report.
The default property file can be used to generate a custom version of properties.rc. It is derived automatically from this documentation, Documentation/pipsmake-rc.tex.
PIPS behavior can also be altered by Shell environment variables. Their generic names is XXXX_DEBUG_LEVEL, where XXXX is a library or a phase or an interface name (of course, there are exceptions). Theoretically these environment variables are also declared as properties, but this is generally forgotten by programmers. A debug level of 0 is equivalent to no tracing. The amount of tracing increases with the debug level. The maximum useful value is 9.
Another Shell environment variable, NEWGEN_MAX_TABULATED_ELEMENTS, is useful to analyze large programs. Its default value is 12,000 but it is not uncommon to have to set it up to 200,000.
Properties are listed below on a source library basis. Properties used in more than one library or used by PIPS infrastructure are presented first. Section 2.3 contains information about properties related to infrastructure, external and user interface libraries. Properties for analyses are grouped in Chapter 6. Properties for program transformations, parallelization and distribution phases are listed in the next section in Chapters 9 and 8. User output produced by different kinds of prettyprinters are presented in Chapter 10. Chaper 11 is dedicated to properties of the libraries added by CEA to implement Feautrier’s method.
Rule and object declaration are grouped in chapters: input files (Chapter 3), syntax analysis and abstract syntax tree (Chapter 4), analyses (Chapter 6), parallelizations (Chapter 8), program transformations (Chapter 9) and prettyprinters of output files (Chapter 10). Chapter 11 describes several analyses defined by Paul Feautrier. Chapter 12 contains a set of menu declarations for the window-based interfaces.
Virtually every PIPS programmer contributed some lines in this report. Inconsistencies are likely. Please report them to the PIPS team9 !
Options are called properties in PIPS. Most of them are related to a specific phase, for instance the dependence graph computation. They are declared next to the corresponding phase declaration. But some are related to one library or even to several libraries and they are declared in this chapter.
Skip this chapter on first reading. Also skip this chapter on second reading because you are unlikely to need these properties until you develop in PIPS.
Are DO loops bodies executed at least once (F-66 style), or not (Fortran 77)?
ONE_TRIP_DO FALSE
is useful for use/def and semantics analysis but is not used for region analyses. This dangerous property should be set to FALSE. It is not consistently checked by PIPS phases, because nobody seems to use this obsolete Fortran feature anymore.
With
LOG_TIMINGS FALSE
it is possible to display the amount of real, cpu and system times directly spent in each phase as well as the times spent reading/writing data structures from/to PIPS database. The computation of total time used to complete a pipsmake request is broken down into global times, a set of phase times which is the accumulation of the times spent in each phase, and a set of IO times, also accumulated through phases.
Note that the IO times are included in the phase times.
With
LOG_MEMORY_USAGE FALSE
it is possible to log the amount of memory used by each phase and by each request. This is mainly useful to check if a computation can be performed on a given machine. This memory log can also be used to track memory leaks. Valgrind may be more useful to track memory leaks.
PIPS infrastructure is based on a few external libraries, Newgen and Linear, and on three key PIPS1 libraries:
Newgen offers some debugging support to check object consistency (gen_consistent_p and gen_defined_p), and for dynamic type checking. See Newgen documentation[?][?].
This library is external and offers an independent debugging system.
The following properties specify how null (
SYSTEM_NULL "<null␣system>"
), undefined
SYSTEM_UNDEFINED "<undefined␣system>"
) or non feasible systems
SYSTEM_NOT_FEASIBLE "{0==-1}"
are prettyprinted by PIPS.
With
CHECK_RESOURCE_USAGE FALSE
it is possible to log and report differences between the set of resources actually read and written by the procedures called by pipsmake and the set of resources declared as read or written in pipsmake.rc file.
ACTIVATE_DEL_DERIVED_RES TRUE
controls the rule activation process that may delete from the database all the derived resources from the newly activated rule to make sure that non-consistent resources cannot be used by accident.
PIPSMAKE_CHECKPOINTS 0
controls how often resources should be saved and freed. 0 means never, and a positive value means every n applications of a rule. This feature was added to allow long big automatic tpips scripts that can coredump and be restarted latter on close to the state before the core. As another side effect, it allows to free the memory and to keep memory consumption as moderate as possible, as opposed to usual tpips runs which keep all memory allocated. Note that it should not be too often saved, because it may last a long time, especially when entities are considered on big workspaces. The frequency may be adapted in a script, rarely at the beginning to more often latter.
Shell environment variables PIPSDBM_DEBUG_LEVEL can be set to ? to check object consistency when they are stored in the database, and to ? to check object consistency when they are stored or retrieved (in case an intermediate phase has corrupted some data structure unwillingly).
You can control what is done when a workspace is closed and resources are saved. The
PIPSDBM_RESOURCES_TO_DELETE "obsolete"
property can be set to to ”obsolete” or to ”all”.
Note that it is not managed from pipsdbm but from pipsmake, which knows what is obsolete or not.
The top-level library is built on top of the pipsmake and pipsdbm libraries to factorize functions useful to build a PIPS user interface or API.
Property
USER_LOG_P TRUE
controls the logging of the session in the database of the current workspace. This log can be processed by PIPS utility logfile2tpips to generate automatically a tpips script which can be used to replay the current PIPS session, workspace by workspace, regardless of the PIPSuser interface used.
Property
ABORT_ON_USER_ERROR FALSE
specifies how user errors impact execution once the error message is printed on stderr: return and go ahead, usually when PIPS is used interactively, or core dump for debugging purposes and for script executions, especially non-regression tests.
Property
MAXIMUM_USER_ERROR 2
specifies the number of user error allowed before the programs brutally aborts.
Property
ACTIVE_PHASES "PRINT_SOURCE␣PRINT_CODE␣PRINT_PARALLELIZED77_CODE␣PRINT_CALL_GRAPH␣PRINT_ICFG␣TRANSFORMERS_INTER_FULL␣INTERPROCEDURAL_SUMMARY_PRECONDITION␣PRECONDITIONS_INTER_FULL␣ATOMIC_CHAINS␣RICE_SEMANTICS_DEPENDENCE_GRAPH␣MAY_REGIONS"
specifies which pipsmake phases should be used when several phases can be used to produce the same resource. This property is used when a workspace is created. A workspace is the database maintained by PIPS to contain all resources defined for a whole application or for the whole set of files used to create it.
Resources that create ambiguities for pipsmake are at least:
This list must be updated according to new rules and new resources declared in this file. Note that no default parser is usually specified in this property, because it is selected automatically according to the source file suffixes when possible.
Until October 2009, the active phases were:
They still are used for the old non-regression tests.
User warnings may be turned off. Definitely, this is not the default option! Most warnings must be read to understand surprising results. This property is used by library misc.
NO_USER_WARNING FALSE
By default, PIPS reports errors generated by system call stat which is used in library pipsdbm to check the time a resource has been written and hence its temporal consistency.
WARNING_ON_STAT_ERROR TRUE
Error messages are also copied in the Warnings file.
The syntactic constraints of C89 have been eased for declarations in C99, where it is possible to intersperse statement declarations within executable statements. This property is used to request C89 compatible code generation.
C89_CODE_GENERATION FALSE
So the default option is to generate C99 code, which may be changed because it is likely to make the code generated by PIPS unparsable by PIPS.
There is no guarantee that each code generation phase is going to comply with this property. It is up to each developper to decide if this global property is to be used or not in his/her local phase.
tpips is one of PIPS user interfaces.
TPIPS_IS_A_SHELL FALSE
controls whether tpips should behave as an extended shell and consider any input command that is not a tpips command a Shell command.
This property is automatically set to TRUE when pyps is running.
PYPS FALSE
An input program is a set of user Fortran 77, Fortran 90 or C source files and a name, called a workspace. The files are looked for in the current directory, then by using the colon-separated PIPS_SRCPATH variable for other directories where they might be found. The first occurrence of the file name in the ordered directories is chosen, which is consistent with PATH and MANPATH behaviour.
The source files are splitted by PIPS at the program initialization phase to produce one PIPS-private source file for each procedure, subroutine or function, and for each block data. A function like fsplit is used and the new files are stored in the workspace, which simply is a UNIX sub-directory of the current directory. These new files have names suffixed by .f.orig.
Since PIPS performs interprocedural analyses, it expects to find a source code file for each procedure or function called. Missing modules can be replaced by stubs, which can be made more or less precise with respect to their effects on formal parameters and global variables. A stub may be empty. Empty stubs can be automatically generated if the code is properly typed (see Section 3.3).
The user source files should not be edited by the user once PIPS has been started because these editions are not going to be taken into account unless a new workspace is created. But their preprocessed copies, the PIPS source files, safely can be edited while running PIPS. The automatic consistency mechanism makes sure that any information displayed to the user is consistent with the current state of the sources files in the workspace. These source files have names terminated by the standard suffix, .f.
New user source files should be automatically and completely re-built when the program is no longer under PIPS control, i.e. when the workspace is closed. An executable application can easily be regenerated after code transformations using the tpips1 interface and requesting the PRINTED_FILE resources for all modules, including compilation units in C:
display PRINTED_FILE[%ALL]
Note that compilation units can be left out with:
display PRINTED_FILE[%ALLFUNC]
In both cases with C source code, the order of modules may be unsuitable for direct recompilation and compilation units should be included anyway, but this is what is done by explicitly requesting the code regeneration as described in § 3.4.
Note that PIPS expects proper ANSI Fortran 77 code. Its parser was not designed to locate syntax errors. It is highly recommended to check source files with a standard Fortran compiler (see Section 3.2) before submitting them to PIPS.
The Fortran 77 files specified as input to PIPS by the user are preprocessed in various ways.
If the PIPS_CHECK_FORTRAN shell environment variable is defined to false or no or 0, the syntax of the source files is not checked by compiling it with a C compiler.If the PIPS_CHECK_FORTRAN shell environment variable is defined to true or yes or 1, the syntax of the file is checked by compiling it with a Fortran 77 compiler. If the PIPS_CHECK_FORTRAN shell environment variable is not defined, the check is performed according to CHECK_FORTRAN_SYNTAX_BEFORE_RUNNING_PIPS 3.2.1.1.
The Fortran compiler is defined by the PIPS_FLINT environment variable. If it is undefined, the default compiler is f77 -c -ansi).
In case of failure, a warning is displayed. Note that if the program cannot be compiled properly with a Fortran compiler, it is likely that many problems will be encountered within PIPS.
The next property also triggers this preliminary syntactic verification.
CHECK_FORTRAN_SYNTAX_BEFORE_RUNNING_PIPS TRUE
PIPS requires source code for all leaves in its visible call graph. By default, a user error is raised by Function initializer if a user request cannot be satisfied because some source code is missing. It also is possible to generate some synthetic code (also known as stubs) and to update the current module list but this is not a very satisfying option because all interprocedural analysis results are going to be wrong. The user should retrieve the generated .f files in the workspace, under the Tmp directory, and add some assignments (def ) and uses to mimic the action of the real code to have a sufficient behavior from the point of view of the analysis or transformations you want to apply on the whole program. The user modified synthetic files should then be saved and used to generate a new workspace.
If PREPROCESSOR_MISSING_FILE_HANDLING 3.2.1.1 is set to "query", a script can optionally be set to handle the interactive request using PREPROCESSOR_MISSING_FILE_GENERATOR 3.2.1.1. This script is passed the function name and prints the filename on standard output. When empty, it uses an internal one.
Valid settings: error or generate or query.
PREPROCESSOR_MISSING_FILE_HANDLING "error"
PREPROCESSOR_MISSING_FILE_GENERATOR ""
The generated stub can have various default effect, say to prevent over-optimistic parallelization.
STUB_MEMORY_BARRIER FALSE
STUB_IO_BARRIER FALSE
If the file suffix is .F then the file is preprocessed. By default PIPS uses gfortran -E for Fortran files. This preprocessor can be changed by setting the PIPS_FPP environment variable.
Moreover the default preprocessing options are -P -D__PIPS__ -D__HPFC__ and they can be extended (not replaced...) with the PIPS_FPP_FLAGS environment variable.
The file is then split into one file per module using a PIPS specialized version of fsplit2 . This preprocessing also handles
The output of this phase is a set of .f_initial files in per-module subdirectories. They constitute the resource INITIAL_FILE.
A second step of preprocessing is performed to produce SOURCE_FILE files with standard Fortran suffix .f from the .f_initial files. The two preprocessing steps are shown in Figure 3.1.
Each module source file is then processed by top-level to handle Fortran include and to comment out IMPLICIT NONE which are not managed by PIPS. Also this phase performs some transformations of complex constants to help the PIPS parser. Files referenced in Fortran include statements are looked for from the directory where the Fortran file is. The Shell variable PIPS_CPP_FLAGS is not used to locate these include files.
The C preprocessor is applied before the splitting. By default PIPS uses cpp -C for C files. This preprocessor can be changed by setting the PIPS_CPP environment variable.
Moreover the -D__PIPS__ -D__HPFC__ -U__GNUC__ preprocessing options are used and can be extended (not replaced) with the PIPS_CPP_FLAGS environment variable.
This PIPS_CPP_FLAGS variable can also be used to locate the include files. Directories to search are specified with the -Ifile option, as usual for the C preprocessor.
If the PIPS_CHECK_C shell environment variable is defined to false or no or 0, the syntax of the source files is not checked by compiling it with a C compiler. If the PIPS_CHECK_C shell environment variable is defined to true or yes or 1, the syntax of the file is checked by compiling it with a C compiler. If the PIPS_CHECK_C shell environment variable is not defined, the check is performed according to CHECK_C_SYNTAX_BEFORE_RUNNING_PIPS 3.2.2.1.
The environment variable PIPS_CC is used to define the C compiler available. If it is undefined, the compiler chosen is gcc -c ).
In case of failure, a warning is displayed.
If the environement variable PIPS_CPP_FLAGS is defined, it should contain the options -Wall and -Werror for the check to be effective.
The next property also triggers this preliminary syntactic verification.
CHECK_C_SYNTAX_BEFORE_RUNNING_PIPS TRUE
Although its default value is FALSE, it is much safer to set it to true when dealing with new sources files. PIPS is not designed to process non-standard source code. Bugs in source files are not well explained or localized. They can result in weird behaviors and inexpected core dumps. Before complaining about PIPS, it is higly recommended to set this property to TRUE.
Note: the C and Fortran syntactic verifications could be controlled by a unique property.
The Fortran 90 parser is a separate program, derived from gcc Fortran parser. It is activated directly when the workspace is created, and not by pipsmake.
The source files may be placed in different directories and have the same name, which makes resource management more difficult. The default option is to assume that no file name conflicts occur. This is the historical option and it leads to much simpler module names.
PREPROCESSOR_FILE_NAME_CONFLICT_HANDLING FALSE
A source_file contains the code of exactly one module. Source files are created from user source files at program initialization by fsplit or a similar function if fsplit is not available (see Section 3.2). A source file may be updated by the user4 , but not by PIPS. Program transformations are performed on the internal representation (see 4) and visible in the prettyprinted output (see 10).
Source code splitting and preprocessing, e.g. cpp, are performed by the function create_workspace() from the top-level library, in collaboration with db_create_workspace() from library pipsdbm which creates the workspace directory. The user source files have names suffixed by .f or .F if cpp must be applied. They are split into original user_files with suffix .f.orig. These so-called original user files are in fact copies stored in the workspace. The syntactic PIPS preprocessor is applied to generate what is known as a source_file by PIPS. This process is fully automatized and not visible from PIPS user interfaces. However, the cpp preprocessor actions can be controlled using the Shell environment variable PIPS_CPP_FLAGS.
Function initializer is only called when the source code is not found. If the user code is properly typed, it is possible to force initializer to generate empty stubs by setting properties PREPROCESSOR_MISSING_FILE_HANDLING 3.2.1.1 and, to avoid inconsistency, PARSER_TYPE_CHECK_CALL_SITES 4.2.1.4. But remember that many Fortran codes use subroutines with variable numbers of arguments and with polymorphic types. Fortran varargs mechanism can be achieved by using or not the second argument according to the first one. Polymorphism can be useful to design an IO package or generic array subroutine, e.g. a subroutine setting an array to zero or a subroutine to copy an array into another one.
The current default option is to generate a user error if some source code is missing. This decision was made for two reasons:
Sometimes, a function happen to be defined (and not only declared) inside a header file with the inline keyword. In that case PIPS can consider it as a regular module or just ignore it, as its presence may be system-dependant. Property IGNORE_FUNCTION_IN_HEADER 3.3 control this behavior and must be set before workspace creation.
IGNORE_FUNCTION_IN_HEADER TRUE
Modules can be flagged as “stubs”, aka functions provided to PIPS but which shouldn’t be inlined or modified. Property PREPROCESSOR_INITIALIZER_FLAG_AS_STUB 3.3 controls if the initializer should declare new files as stubs.
PREPROCESSOR_INITIALIZER_FLAG_AS_STUB TRUE
Note: the generation of the resource user_file here above is mainly directed in having the resource concept here. More thought is needed to have the concept of user files managed by pipsmake.
MUST appear after initializer:
In C, the initializer can generate directly a c_source_file and its compilation unit.
The unsplit 3.4 phase regenerates user files from available printed_file. The various modules that where initially stored in single file are appended together in a file with the same name. Not that just fsplit is reversed, not a preprocessing through cpp. Also the include file preprocessing is not reversed.
Regeneration of user files. The various modules that where initially stored in single file are appended together in a file with the same name.
The abstract syntax tree, a.k.a intermediate representation, a.k.a. internal representation, is presented in [?] and in PIPS Internal Representation of Fortran and C code1 .
Program entities are stored in PIPS unique symbol table2 , called entities. Fortran entities, like intrinsics and operators, are created by bootstrap at program initialization. The symbol table is updated with user local and global variables when modules are parsed or linked together. This side effect is not disclosed to pipsmake.
The entity data structure is described in PIPS Internal Representation of Fortran and C code3 .
The declaration of new intrinsics is not easy because it was assumed that there number was fixed and limited by the Fortran standard. In fact, Fortran extensions define new ones. To add a new intrinsic, C code in bootstrap/bootstrap.c and in effects-generic/intrinsics.c must be added to declare its name, type and Read/Write memory effects.
Information about entities generated by the parsers is printed out conditionally to property: PARSER_DUMP_SYMBOL_TABLE 4.2.1.4. which is set to false by default. Unless you are debugging the parser, do not set this property to TRUE but display the symbol table file. See Section 4.2.1.4 for Fortran and Section 4.2.3 for C.
Each module source code is parsed to produce an internal representation called parsed_code and a list of called module names, callees.
Source code is assumed to be fully Fortran-77 compliant. The syntax should be checked by a standard Fortran compiler, e.g. fort77 or at least gfortran, before the PIPS Fortran 77 parser is activated. On the first encountered error, the parser may be able to emit a useful message or the non-analyzed part of the source code is printed out.
PIPS input language is standard Fortran 77 with few extensions and some restrictions. The input character set includes underscore, _, and varying length variable names, i.e. they are not restricted to 6 characters are supported as well as dependent types for arrays.
PARSER_EXPAND_STATEMENT_FUNCTIONS 4.2.1.4
is set. If the substitution is considered possibly unsafe, a warning is displayed.
These parser restrictions were based on funding constraints. They are mostly alleviated by the preprocessing phase. PerfectClub and SPEC-CFP95 benchmarks are handled without manual editing, but for ENTRY statements which are obsoleted by the current Fortran standard.
For parser debugging purposes, it is possible to print a summary of the symbol table, when enabling this property:
PARSER_DUMP_SYMBOL_TABLE FALSE
This should be avoided and the resource symbol_table_file be displayed instead.
The prettyprint of the symbol table for a Fortran module is generated with:
Some subtle errors occur because the PIPS parser uses a fixed format. Columns 73 to 80 are ignored, but the parser may emit a warning if some characters are encountered in this comment field.
PARSER_WARN_FOR_COLUMNS_73_80 TRUE
PIPS has been initially developed to parse correct Fortran compliant programs only. Real applications use lots of ANSI extensions… and they are not always correct! To make sure that PIPS output is correct, the input code should be checked against ANSI extensions using property
CHECK_FORTRAN_SYNTAX_BEFORE_PIPS
(see Section 3.2) and the property below should be set to false.
PARSER_ACCEPT_ANSI_EXTENSIONS TRUE
Currently, this property is not used often enough in PIPS parser which let go many mistakes... as expected by real users!
PIPS has been developed to parse correct Fortran-77 compliant programs only. Array ranges are used to improve readability. They can be generated by PIPS prettyprinter. They are not parsed as correct input by default.
PARSER_ACCEPT_ARRAY_RANGE_EXTENSION FALSE
Each argument list at calls to a function or a subroutine is compared to the functional type of the callee. Turn this off if you need to support variable numbers of arguments or if you use overloading and do not want to hear about it. For instance, an IO routine can be used to write an array of integers or an array of reals or an array of complex if the length parameter is appropriate.
Since the functional typing is shaky, let’s turn it off by default!
PARSER_TYPE_CHECK_CALL_SITES FALSE
The PIPS implementation of Allen&Kennedy algorithm cannot cope with labeled DO loops because the loop, and hence its label, may be replicated if the loop is distributed. The parser can generate an extra CONTINUE statement to carry the label and produce a label-free loop. This is not the standard option because PIPS is designed to output code as close as possible to the user source code.
PARSER_SIMPLIFY_LABELLED_LOOPS FALSE
Most PIPS analyses work better if do loop bounds are affine. It is sometimes possible to improve results for non-affine bounds by assigning the bound to an integer variables and by using this variable as bound. But this is implemented for Fortran, but not for C.
PARSER_LINEARIZE_LOOP_BOUNDS FALSE
The entry construct can be seen as an early attempt at object-oriented programming. The same object can be processed by several function. The object is declared as a standard subroutine or function and entry points are placed in the executable code. The entry points have different sets of formal parameters, they may share some common pieces of code, they share the declared variables, especially the static ones.
The entry mechanism is dangerous because of the flow of control between entries. It is now obsolete and is not analyzed directly by PIPS. Instead each entry may be converted into a first class function or subroutine and static variables are gathered in a specific common. This is the default option. If the substitution is not acceptable, the property may be turned off and entries results in a parser error.
PARSER_SUBSTITUTE_ENTRIES TRUE
Alternate returns are put among the obsolete Fortran features by the Fortran 90 standard. It is possible (1) to refuse them (option ”NO”), or (2) to ignore them and to replace alternate returns by STOP (option ”STOP”), or (3) to substitute them by a semantically equivalent code based on return code values (option ”RC” or option ”HRC”). Option (2) is useful if the alternate returns are used to propagate error conditions. Option (3) is useful to understand the impact of the alternate returns on the control flow graph and to maintain the code semantics. Option ”RC” uses an additional parameter while option ”HRC” uses a set of PIPS run-time functions to hide the set and get of the return code which make declaration regeneration less useful. By default, the first option is selected and alternate returns are refused.
To produce an executable code, the declarations must be regenerated: see property PRETTYPRINT_ALL_DECLARATIONS 10.2.22.6 in Section 10.2.22.6. This is not necessary with option ”HRC”. Fewer new declarations are needed if variable PARSER_RETURN_CODE_VARIABLE 4.2.1.4 is implicitly integer because its first letter is in the I-N range.
With option (2), the code can still be executed if alternate returns are used only for errors and if no errors occur. It can also be analyzed to understand what the normal behavior is. For instance, OUT regions are more likely to be exact when exceptions and errors are ignored.
Formal and actual label variables are replaced by string variables to preserve the parameter ordre and as much source information as possible. See PARSER_FORMAL_LABEL_SUBSTITUTE_PREFIX 4.2.1.4 which is used to generate new variable names.
PARSER_SUBSTITUTE_ALTERNATE_RETURNS "NO"
PARSER_RETURN_CODE_VARIABLE "I_PIPS_RETURN_CODE_"
PARSER_FORMAL_LABEL_SUBSTITUTE_PREFIX "FORMAL_RETURN_LABEL_"
The internal representation can be hidden and the alternate returns can be prettyprinted at the call sites and modules declaration by turning on the following property:
PRETTYPRINT_REGENERATE_ALTERNATE_RETURNS FALSE
Using a mixed C / Fortran RI is troublesome for comments handling: sometimes the comment guard is stored in the comment, sometime not. Sometimes it is on purpose, sometimes it is not. When following property is set to true, PIPS4 does its best to prettyprint comments correctly.
PRETTYPRINT_CHECK_COMMENTS TRUE
If all modules have been processed by PIPS, it is possible not to regenerate alternate returns and to use a code close to the internal representation. If they are regenerated in the call sites and module declaration, they are nevertheless not used by the code generated by PIPS which is consistent with the internal representation.
Here is a possible implementation of the two PIPS run-time subroutines required by the hidden return code (”HRC”) option:
subroutine SET_I_PIPS_RETURN_CODE_(irc)
common /PIPS_RETURN_CODE_COMMON/irc_shared
irc_shared = irc
end
subroutine GET_I_PIPS_RETURN_CODE_(irc)
common /PIPS_RETURN_CODE_COMMON/irc_shared
irc = irc_shared
end
Note that the subroutine names depend on the PARSER_RETURN_CODE_VARIABLE 4.2.1.4 property. They are generated by prefixing it with SET_ and GET_. There implementation is free. The common name used should not conflict with application common names. The ENTRY mechanism is not used because it would be desugared by PIPS anyway.
By default, assigned GO TO and ASSIGN statements are not accepted. These constructs are obsolete and will not be part of future Fortran standards.
However, it is possible to replace them automatically in a way similar to computed GO TO. Each ASSIGN statement is replaced by a standard integer assignment. The label is converted to its numerical value. When an assigned GO TO with its optional list of labels is encountered, it is transformed into a sequence of logical IF statement with appropriate tests and GO TO’s. According to Fortran 77 Standard, Section 11.3, Page 11-2, the control variable must be set to one of the labels in the optional list. Hence a STOP statement is generated to interrupt the execution in case this happens, but note that compilers such as SUN f77 and g77 do not check this condition at run-time (it is undecidable statically).
PARSER_SUBSTITUTE_ASSIGNED_GOTO FALSE
Assigned GO TO without the optional list of labels are not processed. In other words, PIPS make the optional list mandatory for substitution. It usually is quite easy to add manually the list of potential targets.
Also, ASSIGN statements cannot be used to define a FORMAT label. If the desugaring option is selected, an illegal program is produced by PIPS parser.
This property controls the processing of Fortran statement functions by text substitution in the parser. No other processing is available and the parser stops with an error message when a statement function declaration is encountered.
The default used to be not to perform this unchecked replacement, which might change the semantics of the program because type coercion is not enforced and actual parameters are not assigned to intermediate variables. However most statement functions do not require these extra-steps and it is legal to perform the textual substitution. For user convenience, the default option is textual substitution.
Note that the parser does not have enough information to check the validity of the transformation, but a warning is issued if legality is doubtful. If strange results are obtained when executing codes transformed with PIPS, his property should be set to false.
A better method would be to represent them somehow a local functions in the internal representation, but the implications for pipsmake and other issues are clearly not all foreseen…(Fabien Coelho).
PARSER_EXPAND_STATEMENT_FUNCTIONS TRUE
This parser takes a different Fortran file but applies the same processing as the previous parser. The Fortran file is the result of the preprocessing by the hpfc_filter 8.3.2.1 phase of the original file in order to extract the directives and switch them to a Fortran 77 parsable form. As another side-effect, this parser hides some callees from pipsmake. This callees are temporary functions used to encode HPF directives. Their call sites are removed from the code before requesting full analyses to PIPS. This parser is triggered automatically by the hpfc_close 8.3.2.5 phase when requested. It should never be selected or activated by hand.
A C file is seen in PIPS as a compilation unit, that contains all the objects declarations that are global to this file, and as many as module (function or procedure) definitions defined in this file.
Thus the compilation unit contains the file-global macros, the include statements, the local and global variable definitions, the type definitions, and the function declarations if any found in the C file.
When the PIPS workspace is created by PIPS preprocessor, each C file is preprocessed5 using for instance gcc -E6 and broken into a new which contains only the file-global variable declarations, the function declarations and the type definitions, and one C file for each C function defined in the initial C file.
The new compilation units must be parsed before the new files, containing each one exactly one function definition, can be parsed. The new compilation units are named like the initial file names but with a bang extension.
For example, considering a C file foo.c with 2 function definitions:
After preprocessing, it leads to a file foo.cpp_processed.c that is then split into a new foo!.cpp_processed.c compilation unit containing
and 2 module files containing the definitions of the 2 functions, a calc.c
and a main.c
Note that it is possible to have an empty compilation unit and no module file if the original file does not contain sensible C informations (such as an empty file containing only blank characters and so on).
The resource COMPILATION_UNIT.declarations produced by compilation_unit_parser is a special resource used to force the parsing of the new compilation unit before the parsing of its associated functions. It is in fact a hash table containing the file-global C keywords and typedef names defined in each compilation unit.
In fact phase compilation_unit_parser also produces parsed_code and callees resources for the compilation unit. This is done to work around the fact that rule c_parser was invoked on compilation units by later phases, in particular for the computation of initial preconditions, breaking the declarations of function prototypes. These two resources are not declared here because pipsmake gets confused between the different rules to compute parsed code : there is no simple way to distinguish between compilation units and modules at some times and handling them similarly at other times.
If you want to parse some C code using tpips, it is possible to select the C parser with
A prettyprint of the symbol table for a C module can be generated with
The EXTENDED_VARIABLE_INFORMATION 4.2.3 property can be used to extend the information available for variables. By default the entity name, the offset and the size are printed. Using this property the type and the user name, which may be different from the internal name, are also displayed.
EXTENDED_VARIABLE_INFORMATION FALSE
The C_PARSER_RETURN_SUBSTITUTION 4.2.3 property can be used to handle properly multiple returns within one function. The current default value is false, which preserves best the source aspect but modifies the control flow because the calls to return are assumed to flow in sequence. If the property is set to true, C return statement are replaced, when necessary, either by a simple goto for void functions, or by an assignment of the returned value to a special variable and a goto. A unique return statement is placed at the syntactic end of the function. For functions with no return statement or with a unique return statement placed at the end of their bodies, this property is useless.
C_PARSER_RETURN_SUBSTITUTION FALSE
The C99 for-loop with a declaration such as for(int i = a;...;...) can be represented in the RI with a naive representation such as:
This is done when the C_PARSER_GENERATE_NAIVE_C99_FOR_LOOP_DECLARATION 4.2.3 property is set to TRUE
C_PARSER_GENERATE_NAIVE_C99_FOR_LOOP_DECLARATION FALSE
Else, we can generate more or less other representation. For example, with some declaration splitting, we can generate a more representative version:
if C_PARSER_GENERATE_COMPACT_C99_FOR_LOOP_DECLARATION 4.2.3 property set to FALSE.
C_PARSER_GENERATE_COMPACT_C99_FOR_LOOP_DECLARATION FALSE
Else, we can generate a more compact (but newer representation that can choke some parts of PIPS7 ...) like:
This representation is not yet implemented.
The Fortran 90 parser is not integrated in pipsmake. It is activated earlier when the workspace is created.
PIPS analyses and transformations take advantage of a hierarchical control flow graph (HCFG), which preserves structured part of code as such, and uses a control flow graph only when no syntactic representation is available (see [?]). The encoding of the relationship between structured and unstructured parts of code is explained elsewhere, mainly in the PIPS Internal Representation of Fortran and C code8 .
The controlizer 4.3 is the historical controlizer phase that removes GOTO statements in the parsed code and generates a similar representation with small CFGs. It was developped for Fortran 77 code.
The Fortran controlizer phase was too hacked and undocumented to be improved and debugged for C99 code so a new version has been developed, documented and is designed to be simpler and easier to understand. But, for comparison, the Fortran controlizer phase can still be used.
For debugging and validation purpose, by setting at most one of the PIPS_USE_OLD_CONTROLIZER or PIPS_USE_NEW_CONTROLIZER environment variables, you can force the use of the specific version of the controlizer you want to use. This override the setting by activateRonan?.
Note that the controlizer choice impacts the HCFG when Fortran entries are used. If you do not know what Fortran entries are, it is deprecated stuff anyway... ☺
The new_controlizer 4.3 removes GOTO statements in the parsed code and generates a similar representation with small CFGs. It is designed to work according to C and C99 standards. Sequences of sequence and variable declarations are handled properly. However, the prettyprinter is tuned for code generated by controlizer 4.3, which does not always minimize the number of goto statements regenerated.
The hierarchical control flow graph built by the controlizer 4.3 is pretty crude. The partial control flow graphs, called unstructured statements, are derived from syntactic constructs. The control scope of an unstructured is the smallest enclosing structured construct, whether a loop, a test or a sequence. Thus some statements, which might be seen as part of structured code, end up as nodes of an unstructured.
Note that sequences of statements are identified as such by controlizer 4.3. Each of them appears as a unique node.
Also, useless CONTINUE statements may be added as provisional landing pads and not removed. The exit node should never have successors but this may happen after some PIPS function calls. The exit node, as well as several other nodes, also may be unreachable. After clean up, there should be no unreachable node or the only unreachable node should be the exit node. Function unspaghettify 9.3.3.1 (see Section 9.3.3.1) is applied by default to clean up and to reduce the control flow graphs after controlizer 4.3.
The GOTO statements are transformed in arcs but also in CONTINUE statements to preserve as many user comments as possible.
The top statement of a module returned by the controlizer 4.3 used to contain always an unstructured instruction with only one node. Several phases in PIPS assumed that this always is the case, although other program transformations may well return any kind of top statement, most likely a block. This is no longer true. The top statement of a module may contain any kind of instruction.
Here is declared the C and C99 controlizer:
Control restructuring eliminates empty sequences but as empty true or false branch of structured IF. This semantic property of PIPS Internal Representation of Fortran and C code9 is enforced by libraries effects, regions, hpfc, effects-generic.
WARN_ABOUT_EMPTY_SEQUENCES FALSE
By unsetting this property unspaghettify 9.3.3.1 is not applied implicitly in the controlizer phase.
UNSPAGHETTIFY_IN_CONTROLIZER TRUE
The next property is used to convert C for loops into C while loops. The purpose is to speed up the re-use of Fortran analyses and transformation for C code. This property is set to false by default and should ultimately disappear. But for new user convenience, it is set to TRUE by activate_language() when the language is C.
FOR_TO_WHILE_LOOP_IN_CONTROLIZER FALSE
The next property is used to convert C for loops into C do loops when syntactically possible. The conversion is not safe because the effect of the loop body on the loop index is not checked. The purpose is to speed up the re-use of Fortran analyses and transformation for C code. This property is set to false by default and should disappear soon. But for new user convenience, it is set to TRUE by activate_language() when the language is C.
FOR_TO_DO_LOOP_IN_CONTROLIZER FALSE
This can also explicitly applied by calling the phase described in § 9.3.3.4.
To able deeper code transformation, FORMATs can be gathered at the very beginning of the code or at the very end according to the following options in the unspaghettify or control restructuring phase.
GATHER_FORMATS_AT_BEGINNING FALSE
GATHER_FORMATS_AT_END FALSE
To display the statistics about cleaning-up sequences and removing useless CONTINUE or empty statement.
CLEAN_UP_SEQUENCES_DISPLAY_STATISTICS FALSE
There is a trade-off between keeping the comments associated to labels and goto and the cleaning that can be do on the control graph.
By default, do not fuse empty control nodes that have labels or comments:
FUSE_CONTROL_NODES_WITH_COMMENTS_OR_LABEL FALSE
By default, do not fuse sequences with internal declarations. Turning this to TRUE results in variable renamings when the same variable name is used at several places in the analyzed module.
CLEAN_UP_SEQUENCES_WITH_DECLARATIONS FALSE
The internal representation includes special field to declare parallel constructs such as parallel loops. A parallel code internal representation does not differ fundamentally from a sequential code.
Although this phases should be spread elsewhere in this manual, we have put some pedagogical phases useful to jump into PIPS first.
A phase that displays, in debug mode, statements matching an XPath expression on the internal representation:
Prepends a comment to the first statement of a module. Useful to apply post-processing after PIPS.
The comment to add is selected by this property:
PREPEND_COMMENT "/*␣This␣comment␣is␣added␣by␣PREPEND_COMMENT␣phase␣*/"
This phase inserts a call to function MY_TRACK just before the first statement of a module. Useful as a pedagogical example to explore the internal representation and Newgen. Not to be used for any pratical purpose as it is bugged. Debugging it is a pedagogical exercise.
The called function could be defined by this property:
PREPEND_CALL "MY_TRACK"
but it is not.
This phase prepend or appends a pragma to a module.
The pragma name can be defined by this property:
PRAGMA_NAME "MY_PRAGMA"
The pragma can be append or prepend thanks to this property:
PRAGMA_PREPEND TRUE
The pass clear_pragma 5.4 clears all pragma, this should be done on any input with unhandled pragma, we don’t what semantic we might break.
The pass pragma_outliner 5.4 is used for outlining a sequence of statements contained between two given sentinel pragmas using properties PRAGMA_OUTLINER_BEGIN 5.4 and PRAGMA_OUTLINER_END 5.4. The name of the new function is controlled using PRAGMA_OUTLINER_PREFIX 5.4.
PRAGMA_OUTLINER_BEGIN "begin"
PRAGMA_OUTLINER_END "end"
PRAGMA_OUTLINER_PREFIX "pips_outlined"
Remove labels that are not usefull
Loop labels can be kept thanks to this property:
REMOVE_USELESS_LABEL_KEEP_LOOP_LABEL FALSE
Analyses encompass the computations of call graphs, the memory effects, reductions, use-def chains, dependence graphs, interprocedural checks (flinter), semantics information (transformers and preconditions), continuations, complexities, convex array regions, dynamic aliases and complementary regions.
All lists of callees are needed to build the global lists of callers for each module. The callers and callees lists are used by pipsmake to control top-down and bottom-up analyses. The call graph is assumed to be a DAG, i.e. no recursive cycle exists, but it is not necessarily connected.
The height of a module can be used to schedule bottom-up analyses. It is zero if the module has no callees. Else, it is the maximal height of the callees plus one.
The depth of a module can be used to schedule top-down analyses. It is zero if the module has no callers. Else, it it the maximal depth of the callers plus one.
The following pass generates a uDrawGraph1 version of the callgraph. Its quite partial since it should rely on an hypothetical all callees, direct and indirect, resource.
The data structures used to represent memory effects and their computation are described in [?]. Another description is available on line, in PIPS Internal Representation of Fortran and C code2 Technical Report.
Note that the standard name in the Dragon book is likely to be Gen and Kill sets in the standard data flow analysis framework, but PIPS uses the more general concept of effect developped by P. Jouvelot and D. Gifford [?] and its analyses are mostly based on the abstract syntac tree (AST) rather than the control flow graph (CFG).
The proper memory effects of a statement basically are a list of variables that may be read or written by the statement. They are used to build use-def chains (see [?] or a later edition) and then the dependence graph.
Proper means that the effects of a compound statement do not include the effects of lower level statements. For instance, the body of a loop, true and false branches of a test statement, control nodes in an unstructured statement ... are ignored to compute the proper effects of a loop, a test or an unstructured.
Two families of effects are computed : pointer_effects are effects in which intermediary access paths may refer to different memory locations at different program points; regular effects are constant path effects, which means that their intermediary access paths all refer to unique memory locations. The same distinction holds for convex array regions (see section 6.12).
proper_effects_with_points_to and proper_effects_with_pointer_values are alternatives to compute constant path proper effects using points-to (see subsection 6.13.3) or pointer values analyses (see subsection 6.13.6). This is still at an experimental stage.
Summary effects (see Section 6.2.4) of a called module are used to compute the proper effects at the corresponding call sites. They are translated from the callee’s scope into the caller’s scope. The translation is based on the actual-to-formal binding. If too many actual arguments are defined, a user warning is issued but the processing goes on because a simple semantics is available: ignore useless actual arguments. If too few actual arguments are provided, a user error is issued because the effects of the call are not defined.
Variables private to loops are handled like regular variable.
See proper_effects 6.2.1
See proper_effects 6.2.1
To be continued...by whom?
Cumulated effects of statements are lists of read or written variables, just like the proper effects (see Section 6.2.1). Cumulated means that the effects of a compound statement, do loop, test or unstructured, include the effects of the lower level statements such as a loop body or a test branch.
Summary data flow information is the simplest interprocedural information needed to take procedure into account in a parallelizer. It was introduced in Parafrase (see [?]) under this name, but should be called summary memory effects in PIPS context.
The summary_effects 6.2.4 of a module are the cumulated memory effects of its top level statement (see Section 6.2.3), but effects on local dynamic variables are ignored (because they cannot be observed by the callers3 ) and subscript expressions of remaining effects are eliminated.
IN and OUT memory effects of a statement s are memory locations whose input values are used by statement s or whose output values are used by statement s continuation. Variables allocated in the statement are not part of the IN or OUT effects. Variables defined before they are used ar not part of the IN effects. OUT effects require an interprocedural analysis4
The concept of proper references is not yet clearly defined. The original idea is to keep track of the actual objects of Newgen domain reference used in the program representation of the current statement, while retaining if they correspond to a read or a write of the corresponding memory locations. Proper references are represented as effects.
For C programs, where memory accesses are not necessarily represented by objects of Newgen domain reference, the semantics of this analysis is unclear.
Cumulated references gather proper references over the program code, without taking into account the modification of memory stores by the program execution.
FC: I should implement real summary references?
Filter this variable in phase filter_proper_effects 6.2.2.
EFFECTS_FILTER_ON_VARIABLE ""
When set to TRUE, EFFECTS_POINTER_MODIFICATION_CHECKING 6.2.7 enables pointer modification checking during the computation of cumulated effects and/or RW covex array regions. Since this is still at experimentation level, it’s default value is FALSE. This property should disappear when pointer modification analyses are more mature.
EFFECTS_POINTER_MODIFICATION_CHECKING FALSE
The default (and correct) behaviour for the computation of effects is to transform dereferencing paths into constant paths. When property CONSTANT_PATH_EFFECTS 6.2.7 is set to FALSE, the latter transformation is skipped. Effects are then equivalent to pointer_effects. This property is available for backward compatibility and experimental purpose. It must be borne in mind that analyses and transformations using the resulting effects may yield uncorrect results. This property also affects the computation of convex array regions.
CONSTANT_PATH_EFFECTS TRUE
Since CONSTANT_PATH_EFFECTS 6.2.7 may be set to FALSE erroneously, some tests are included in conflicts testing to avoid generating wrong code. However, these tests are costly, and can be turned off by setting TRUST_CONSTANT_PATH_EFFECTS_IN_CONFLICTS 6.2.7 to FALSE. This must be used with care and only when there is no aliasing.
TRUST_CONSTANT_PATH_EFFECTS_IN_CONFLICTS FALSE
Property USER_EFFECTS_ON_STD_FILES 6.2.7 is used to control the way the user uses stdout, stdin and stderr. The default case (FALSE) means that the user does not modify these global variables. When set to TRUE, they are considered as user variables, and dereferencing them through calls to stdio functions leads to less precise effects.
USER_EFFECTS_ON_STD_FILES FALSE
Property MEMORY_EFFECTS_ONLY 6.2.7 is used to restrict the action kind of an effect action to store. In other words, variable declarations and type declarations are not considered to alter the execution state when this property is set to TRUE. This is fine for Fortran code because variables cannot be declared among executable statements and because new type cannot be declared. But this leads to wrong result for C code when loop distribution or use-def elimination is performed.
Currently, PIPS does not have the capability to store default values depending on the source code language. The default value is TRUE to avoid disturbing too many phases of PIPS at the same time while environment and type declaration effects are introduced.
MEMORY_EFFECTS_ONLY TRUE
Some programs do measure execution times. All code placed between measurement points must not be moved out, as can happen when loops are distributed or, more generally, instructions are rescheduled. Since loops using time effects are not parallel, a clock variable is always updated when a time-related function is called. This is sufficient to avoid most problems, but not all of them because time effects of all other executed statements are kept implicit, i.e. the real time clock is not updated: and loops can still be distributed. If time measurements are key, this property must be turned on. By default, it is turned off.
TIME_EFFECTS_USED FALSE
Some source code is sometimes missing. PIPS5 does not have any way to guess the memory effects of functions whose source code is missing. Several approaches are possible to approximate the exact effects. Two optimistic ones are implemented: either we assume that the function only computes a result and has no side effects thru pointer parameters, global variables or static variables (default option), or we assume the maximal possible effects through pointers (this should be clarified: for all pointers p, *p is written) but not thru static or global variables.
MAXIMAL_PARAMETER_EFFECTS_FOR_UNKNOWN_FUNCTIONS FALSE
For safety, a pessimitic option is be implemented and a maximal memory effect, *ANYMODULE*:*ANYWHERE*, is associated to such unknown functions.
MAXIMAL_EFFECTS_FOR_UNKNOWN_FUNCTIONS FALSE
These two properties should not be true simultaneously.
There are many cases in which it is necessary to know if a variable may be used in the remainder of the execution of the analyzed application. For instance, a global variable cannot be privatized, or a global array be scalarized, if we don’t know whether their values are used afterwards, or a copy-out must be generated, which is not currently implemented in the simplest algorithms. Similarly, preconditions do not need to propagate information about variables which are no more alive.
Traditional liveness analyzes deals with scalar variables. However, with C code, it is interesting to be able to distinguish between different structure fields for instance, and it may also be interesting to deal with array regions. So we have retained for these analyzes an internal representation as effects, thus allowing to deal with general memory access paths, and to rely on the existing machinery of effects/regions computations.
For each statement or function, we compute two sets: a Live_in set contains the memory paths which are alive in the store preceding the statement execution, while a Live_out set contains the memory paths alive in the store immediately following the statement execution. For sequences of instructions, the Live_out set of an instruction is equal to the Live_in set of the next instruction. However, this not true for the last statement of conditional or loop bodies and the first statement of the next instruction.
The proper reductions are computed from a code.
The cumulated reductions propagate the reductions in the code, upwards.
This pass summarizes the reductions candidates found in a module for export to its callers. The summary effects should be used to restrict attention to variable of interest in the translation?
Some possible (simple) transformations could be added to the code to mark reductions in loops, for latter use in the parallelization.
The following is NOT implemented. Anyway, should the cumulated_reductions be simply used by the prettyprinter instead?
tries to transform
into
tries to transform
which hides a reduction on b into
when possible
Use-def and def-use chains are a standard data structure in optimizing compilers [?]. These chains are used as a first approximation of the dependence graph. Chains based on convex array regions (see Section 6.12) are more effective for interprocedural parallelization.
If chains based on convex array regions have been selected, the simplest dependence test must be used because regions carry more information than any kind of preconditions. Preconditions and loop bound information already are included in the region predicate.
The algorithm used to compute use-def chains is original because it is based on PIPS hierarchical control flow graph and not on a unique control flow graph.
This algorithm generates inexistent dependencies on loop indices. These dependence arcs appear between DO loop headers and implicit DO loops in IO statements, or between one DO loop header and unrelated DO loop bound expressions using that index variable. It is easy to spot the problem because loop indices are not privatized. A prettyprint option,
PRETTYPRINT_ALL_PRIVATE_VARIABLES 10.2.22.5.1
must be set to true to see if the loop index is privatized or not. The problem disappears when some loop indices are renamed.
The problem is due to the internal representation of DO loops: PIPS has no way to distinguish between initialization effects and increment effects. They have to be merged as proper loop effects. To reduce the problem, proper effects of DO loops do not include the index read effect due to the loop incrementation.
Artificial arcs are added to... (Pierre Jouvelot, help!).
Such chains are required for effective interprocedural parallelization. The dependence graph is annotated with proper regions, to avoid inaccuracy due to summarization at simple statement level (see Section 6.12).
Region-based chains are only compatible with the Rice Fast Dependence Graph option (see Section 6.6.1) which has been extended to deal with them6 . Other dependence tests do not use region descriptors (their convex system), because they cannot improve the Rice Fast Dependence test based on regions.
Beware : this option is for experimental use only; resulting parallel code may not be equivalent to input code (see the explanations below).
When in_out_regions_chains 6.5.4 is selected, IN and OUT regions (see Sections 6.12.5 and 6.12.8) are used at call sites instead of READ and WRITE regions. For all other statements, usual READ and WRITE regions are used.
As a consequence, arrays and scalars which could be declared as local in callees, but are exposed to callers because they are statically allocated or are formal parameters, are ignored, increasing the opportunities to detect parallel loops. But as the program transformation which consists in privatizing variables in modules is not yet implemented in PIPS, the code resulting from the parallelization with in_out_regions_chains 6.5.4 may not be equivalent to the original sequential code. The privatization here is non-standard: for instance, variables declared in commons or static should be stack allocated to avoid conflicts.
As for region-based chains (see Section 6.5.3), the simplest dependence test should be selected for best results.
The following loop in Subroutine inout cannot be parallelized legally because Subroutine foo uses a static variable, y. However, PIPS will display this loop as (potentially) parallel if the in_out option is selected for use-def chain computation. Remember that IN/OUT regions require MUST regions to obtain interesting results (see Section 6.12.5).
subroutine inout(a,n)
real a(n)
do i = 1, n
call foo(a(i))
enddo
end
subroutine foo(x)
save y
y = x
x = x + y
end
It is possible to put use-use dependence arcs in the dependence graph. This is useful for estimation of cache memory traffic and of communication for distributed memory machine (e.g. you can parallelize only communication free loops). Beware of use-use dependence on scalar variables. You might expect scalars to be broadcasted and/or replicated on each processor but they are not handled that way by the parallelization process unless you manage to have them declared private with respect to all enclosing loops.
This feature is not supported by PIPS user interfaces. Results may be hard to interpret. It is useful to print the dependence graph.
KEEP_READ_READ_DEPENDENCE FALSE
It is possible to mask effects on local variables in loop bodies. This is dangerous with current version of Allen & Kennedy which assumes that all the edges are present, the ones on private variables being partially ignored but for loop distribution. In other words, this property should always be set to false.
CHAINS_MASK_EFFECTS FALSE
It also is possible to keep only true data-flow (Def – Use) dependences in the dependence graph. This was an attempt at mimicking the effect of direct dependence analysis and at avoiding privatization. However, direct dependence analysis is not implemented in the standard tests and spurious def-use dependence arcs are taken into account.
CHAINS_DATAFLOW_DEPENDENCE_ONLY FALSE
These last two properties are not consistent with PIPS current development (1995/96). It is assumed that all dependence arcs are present in the dependence graph. Phases using the latter should be able to filter out irrelevant arcs, e.g. pertaining to privatized variables.
The dependence graph is used primarily by the parallelization algorithms. A dependence graph is a refinement of use-def chains (Section 6.5). It is location-based and not value-based.
There are several ways to compute a dependence graph. Some of them are fast (Banerjee’s one for instance) but provide poor results, others might be slower (Rï¿œmi Triolet’s one for instance) but produce better results.
Three different dependence tests are available, all based on Fourier-Motzkin elimination improved with a heuristics for the integer domain. The fast version uses subscript expressions only (unless convex array regions were used to compute use-def chains, in which case regions are used instead). The full version uses subscript expressions and loop bounds. The semantics version uses subscript expressions and preconditions (see 6.9).
Note that, for interprocedural parallelization, precise array regions only are used by the fast dependence test if the proper kind of use-def chains has been previously selected (see Section 6.5.3).
There are several kinds of dependence graphs. Most of them share the same overall data structure: a graph with labels on arcs and vertices. usually, the main differences are in the labels that decorate arcs; for instance, Kennedy’s algorithm requires dependence levels (which loop actually creates the dependence) while algorithms originated from CSRD prefer DDVs (relations between loop indices when the dependence occurs). Dependence cones introduced in [?, ?, ?, ?] are even more precise [?].
The computations of dependence level and dependence cone [?] are both implemented in PIPS. DDV’s are not computed. Currently, only dependence levels are exploited by parallelization algorithms.
The dependence graph can be printed with or without filters (see Section 10.8). The standard dependence graph includes all arcs taken into account by the parallelization process (Allen & Kennedy [?]), except those that are due to scalar private variables and that impact the distribution process only. The loop carried dependence graph does not include intra-iteration dependences and is a good basis for iteration scheduling. The whole graph includes all arcs, but input dependence arcs.
It is possible to gather some statistics about dependences by turning on property RICEDG_PROVIDE_STATISTICS 6.6.6.2 (more details in the properties). A Shell script from PIPS utilities, print-dg-statistics, can be used in combination to extract the most relevant information for a whole program.
During the parallelization phases, is is possible to ignore arcs related to states of the libc, such as the heap memory management, because thread-safe libraries do perform the updates within critical sections. But these arcs are part of the use-def chains and of the dependence graph. If they were removed instead of being ignored, use-def elimination would remove all free statements.
The main contributors for the design and development of dependence analysis are Rï¿œmi Triolet, Franï¿œois Irigoin and Yi-qing Yang [?]. The code was improved by Corinne Ancourt and Bï¿œatrice Creusillet.
Use subscript expressions only, unless convex array regions were used to compute use-def chains, in which case regions are used instead. rice_regions_dependence_graph is a synonym for this rule, but emits a warning if region_chains is not selected.
Use subscript expressions and loop bounds.
Uses subscript expressions and preconditions (see 6.9).
Synonym for rice_fast_dependence_graph, except that it emits a warning when region_chains is not selected.
This property seems to be now obsolete. The dependence test choice is now controlled directly and only by rules in pipsmake. The procedures called by these rules may use this property. Anyway, it is useless to set it manually.
DEPENDENCE_TEST "full"
Provide the following counts during the dependence test. There are three parts: numbers of dependencies and independences (fields 1-10), dimensions of referenced arrays and dependence natures (fields 11-25) and the same information for constant dependencies (fields 26-40), decomposition of the dependence test in elementary steps (fields 41-49), use and complexity of Fourier-Motzkin’s pair-wise elimination (fields 50, 51 and 52-68).
Note: field 1 minus field 2 is the number of array dependencies.
Note: field 5 must be greater or equal to field 4.
Note: the sum of fields 5 to 8 and field 2 equals field 1
Note: the sum of fields 11 to 25 should be equal to the sum of field 9 and 2 minus field 1.
Note: the fields 26 to 40 must be less than or equal to the corresponding fields 11 to 25
Note: the sum of fields 41 to 49 equals field 2
The results are stored in the current workspace in MODULE.resulttestfast, MODULE.resultesttestfull, or MODULE.resulttestseman according to the test selected.
RICEDG_PROVIDE_STATISTICS FALSE
Provide the statistics above and count all array reference pairs including these involved in call statement.
RICEDG_STATISTICS_ALL_ARRAYS FALSE
Only take into account true flow dependences (Def – Use) during the computation of SCC? Note that this is different from the CHAINS_DATAFLOW_DEPENDENCE_ONLY option which doesn’t compute the whole graph. Warning: this option potentially yields incorrect parallel code.
RICE_DATAFLOW_DEPENDENCE_ONLY FALSE
Here are the properties used to control the printing of dependence graphs in a file called module_name.dg. These properties should not be used explicitly because they are set implicitly by the different print-out procedures available in pipsmake.rc. However, not all combinations are available from pipsmake.rc.
PRINT_DEPENDENCE_GRAPH FALSE
To print the dependence graph without the dependences on privatized variables
PRINT_DEPENDENCE_GRAPH_WITHOUT_PRIVATIZED_DEPS FALSE
To print the dependence graph without the non-loop-carried dependences:
PRINT_DEPENDENCE_GRAPH_WITHOUT_NOLOOPCARRIED_DEPS FALSE
To print the dependence graph with the dependence cones:
PRINT_DEPENDENCE_GRAPH_WITH_DEPENDENCE_CONES FALSE
To print the dependence graph in a computer friendly format defined by Deborah Whitfield (SRU):
PRINT_DEPENDENCE_GRAPH_USING_SRU_FORMAT FALSE
The default option is to compute the dependence graph only for loops which can be parallelized using Allen & Kennedy algorithm. However it is possible to compute the dependences in all cases, even for loop containing test, goto, etc... by setting this option to TRUE.
Of course, this information is not used by the parallelization phase which is restricted to loops meeting the A&K conditions. By the way, the hierarchical control flow graph is not exploited either by the parallelization phase.
COMPUTE_ALL_DEPENDENCES FALSE
Function flinter 6.7 performs some intra and interprocedural checks about formal/actual argument pairs, use of COMMONs,... It was developed by Laurent Aniort and Fabien Coelho. Ronan Keryell added the uninitialized variable checking.
In the past, flinter 6.7 used to require MODULE.summary_effects to check the parameter passing modes and to make sure that no module would attempt an assignment on an expression. However, this kind of bug is detected by the effect analysis… which was required by flinter.
Resource CALLEES.code is not explicitly required but it produces the global symbols which function flinter 6.7 needs to check parameter lists.
Computes statistics about loops in module. It computes the number of perfectly and imperfectly nested loops and gives their depths. And it gives the number of nested loops which we can treat with our algorithm.
Note: it does not seem to behave like a standard analysis, associating information to the internal representation. Instead, an ASCII file seems to be created.
PIPS semantics analysis targets mostly integer scalar variables. It is a two-pass process, with a bottom-up pass computing transformers 6.9.1, and a top-down pass propagating preconditions 6.9.2. Transformers and preconditions are specially powerful case of return and jump functions [?]. They abstract relations between program states with polyhedra and encompass most standard interprocedural constant propagations as well as most interval analyses. It is a powerful relational symbolic analysis.
Unlike [?] their computations are based on PIPS Hierarchical Control Flow Graph and on syntactic constructs instead of a standard flow graph. The best presentation of this part of PIPS is in [?].
A similar analysis is available in Parafrase-2 []. It handles polynomial equations between scalar integer variables. SUIF [] also performs some kind of semantics analysis.
The semantics analysis part of PIPS was designed and developed by Franï¿œois Irigoin.
RK: The following is hard to read without any example for someone that knows nothing about PIPS... FI: do you want to have everything in this documentation?
A transformer is an approximate relation between the symbolic initial values of scalar variables and their values after the execution of a statement, simple or compound (see [?] and [?]). In abstract interpretation terminology, a transformer is an abstract command linking the input abstract state of a statement and its output abstract state.
By default, only integer scalar variables are analyzed, but properties can be set to handle boolean, string and floating point scalar variables7 : SEMANTICS_ANALYZE_SCALAR_INTEGER_VARIABLES 6.9.4.1 SEMANTICS_ANALYZE_SCALAR_BOOLEAN_VARIABLES 6.9.4.1 SEMANTICS_ANALYZE_SCALAR_STRING_VARIABLES 6.9.4.1 SEMANTICS_ANALYZE_SCALAR_FLOAT_VARIABLES 6.9.4.1 SEMANTICS_ANALYZE_SCALAR_COMPLEX_VARIABLES 6.9.4.1
Transformers can be computed intraprocedurally by looking at each function independently or they can be computed interprocedurally starting with the leaves of the call tree8 .
Intraprocedural algorithms use cumulated_effects 6.2.3 to handle procedure calls correctly. In some respect, they are interprocedural since call statements are accepted. Interprocedural algorithms use the summary_transformer 6.9.1.7 of the called procedures.
Fast algorithms use a very primitive non-iterative transitive closure algorithm (two possible versions: flow sensitive or flow insensitive). Full algorithms use a transitive closure algorithm based on vector subspace (i.e. ￜ la Karr [?]) or one based on the discrete derivatives [?, ?]. The iterative fix point algorithm for transformers (i.e. Halbwachs/Cousot [?] is implemented but not used because the results obtained with transitive closure algorithms are faster and up-to-now sufficient. Property SEMANTICS_FIX_POINT_OPERATOR 6.9.4.7 is set to select the transitive closure algorithm used.
Additional information, such as array declarations and array references, can be used to improve transformers. See the property documentation for:
SEMANTICS_TRUST_ARRAY_DECLARATIONS 6.9.4.2 SEMANTICS_TRUST_ARRAY_REFERENCES 6.9.4.2
Within one procedure, the transformers can be computed in forward mode, using precondition information gathered along. Transformers can also be recomputed once the preconditions are available. In both cases, more precise transformers are obtained because the statement can be better modelized using precondition information. For instance, a non-linear expression can turn out to be linear because the values of some variables are numerically known and can be used to simplify the initial expression. See properties:
SEMANTICS_RECOMPUTE_EXPRESSION_TRANSFORMERS 6.9.4.5
SEMANTICS_COMPUTE_TRANSFORMERS_IN_CONTEXT 6.9.4.5
SEMANTICS_RECOMPUTE_FIX_POINTS_WITH_PRECONDITIONS 6.9.4.7
and phase refine_transformers 6.9.1.6.
Unstructured control flow graphs can lead to very long transformer computations, whose results are usually not interesting. Their sizes are limited by two properties:
SEMANTICS_MAX_CFG_SIZE2 6.9.4.4 SEMANTICS_MAX_CFG_SIZE1 6.9.4.4
discussed below.
Default value were set in the early nineties to obtain results fast enough for live demonstrations. They have not been changed to preserve the non-regression tests. However since 2005, processors are fast enough to use the most precise options in all cases.
A transformer map contains a transformer for each statement of a module. It is a mapping from statements to transformers (type statement_mapping, which is not a NewGen file). Transformers maps are stored on and retrieved from disk by pipsdbm.
Build the fast intraprocedural transformers.
Build the improved intraprocedural transformers.
Build the fast interprocedural transformers.
Build the improved interprocedural transformers (This should be used as default option.).
Rebuild the interprocedural transformers using interprocedural preconditions. Intraprocedural preconditions are also used to refine all transformers.
A summary transformer is an interprocedural version of the module statement transformer, obtained by eliminating dynamic local, a.k.a. stack allocated, variables. The filtering is based on the module summary effects. Note: each module has a UNIQUE top-level statement.
A summary_transformer 6.9.1.7 is of Newgen type transformer.
A precondition for a statement s in a module m is a predicate true for every state reachable from the initial state of m, in which s is executed. A precondition is of NewGen type ”transformer” (see PIPS Internal Representation of Fortran and C code9 ) and preconditions is of type statement_mapping.
Option preconditions_intra 6.9.2.5 associates a precondition to each statement, assuming that no information is available at the module entry point.
Inter-procedural preconditions may be computed with intra-procedural transformers but the benefit is not clear. Intra-procedural preconditions may be computed with inter-procedural transformers. This is faster that a full interprocedural analysis because there is no need for a top-down propagation of summary preconditions. This is compatible with code transformations like partial_eval 9.4.2, simplify_control 9.3.1 and dead_code_elimination 9.3.2.
Since these two options for transformer and precondition computations are independent and that transformers_inter_full 6.9.1.5 and preconditions_inter_full 6.9.2.7 must be both (independently) selected to obtain the best possible results. These two options are recommended.
All DATA initializations contribute to the global initial state of the program. The contribution of each module is computed independently. Note that variables statically initialized behave as static variables and are preserved between calls according to Fortran standard. The module initial states are abstracted by an initial precondition based on integer scalar variables only.
Note: To be extended to handle C code. To be extended to handle properly unknown modules.
All initial preconditions, including the initial precondition for the main, are combined to define the program precondition which is an abstraction of the program initial state.
The program precondition can only be used for the initial state of the main procedure. Although it appears below for all interprocedural analyses and it always is computed, it only is used when a main procedure is available.
A summary precondition is of type ”transformer”, but the argument list must be empty as it is a simple predicate on the initial state. So in fact it is a state predicate.
The intraprocedural summary precondition uses DATA statement for the main module and is the TRUE constant for all other modules.
Interprocedural summary preconditions can be requested instead. They are not described in the same section in order to introduce the summary precondition resource at the right place in pipsmake.rc.
No menu is declared to select either intra- or interprocedural summary preconditions.
By default, summary preconditions are computed intraprocedurally. The interprocedural option must be explicitly activated.
An interprocedural summary precondition for a module is derived from all its call sites. Of course, preconditions must be known for all its callers’ statements. The summary precondition is the convex hull of all call sites preconditions, translated into a proper environment which is not necessarily the module’s frame. Because of invisible global and static variables and aliasing, it is difficult for a caller to know which variables might be used by the caller to represent a given memory location. To avoid the problem, the current summary precondition is always translated into the caller’s frame. So each module must first translate its summary precondition, when receiving it from the resource manager (pipsdbm) before using it.
Note: the previous algorithm was based on a on-the-fly reduction by convex hull. Each time a call site was encountered while computing a module preconditions, the callee’s summary precondition was updated. This old scheme was more efficient but not compatible with program transformations because it was impossible to know when the summary preconditions of the modules had to be reset to the infeasible (a.k.a. empty) precondition.
An infeasible precondition means that the module is never called although a main is present in the workspace. If no main module is available, a TRUE precondition is generated. Note that, in both cases, the impact of static initializations propagated by link edition is taken into account although this is prohibited by the Fortran Standard which requires a BLOCKDATA construct for such initializations. In other words, a module which is never called has an impact on the program execution and its declarations should not be destroyed.
The following rule is obsolete. It is context sensitive and its results depends on the history of commands performed on the workspace.
Only build the preconditions in a module without any interprocedural propagation. The fast version uses a fast but crude approximation of preconditions for unstructured code.
Option preconditions_inter_fast 6.9.2.6 uses the module own precondition derived from its callers as initial state value and propagates it downwards in the module statement.
The fast versions use no fix-point operations for loops.
Option preconditions_inter_full 6.9.2.7 uses the module own precondition derived from its callers as initial state value and propagates it downwards in the module statement.
The full versions use fix-point operations for loops.
Total preconditions are interesting to optimize the nominal behavior of a terminating application. It is assumed that the application ends in the main procedure. All other exits, aborts or stops, explicit or implicit such as buffer overflows and zero divide and null pointer dereferencing, are considered exceptions. This also applies at the module level. Modules nominally return. Other control flows are considered exceptions. Non-terminating modules have an empty total precondition10 . The standard preconditions can be refined by anding with the total preconditions to get information about the nominal behavior. Similar sources of increased accuracy are the array declarations and the array references, which can be exploited directly with properties described in section 6.9.4.2. These two properties should be set to true whenever possible.
Hence, a total precondition for a statement s in a module m is a predicate true for every state from which the final state of m, in which s is executed, is reached. It is an over-approximation of the theoretical total precondition. So, if the predicate is false, the final control state cannot be reached. A total precondition is of NewGen type ”transformer” (see PIPS Internal Representation of Fortran and C code11 ) and total_preconditions is of type statement_mapping.
The relationship with continuations (see Section 6.10) is not clear. Total preconditions should be more general but no must version exist.
Option total_preconditions_intra 6.9.3.2 associates a precondition to each statement, assuming that no information is available at the module return point.
Inter-procedural total preconditions may be computed with intra-procedural transformers but the benefit is not clear. Intra-procedural total preconditions may be computed with inter-procedural transformers. This is faster than a full interprocedural analysis because there is no need for a top-down propagation of summary total postconditions.
Since these two options for transformer and total precondition computations are independent, transformers_inter_full 6.9.1.5 and total_preconditions_inter 6.9.3.3 must be both (independently) selected to obtain the best possible results.
Status: This is a set of experimental passes. The intraprocedural part is implemented. The interprocedural part is not implemented yet, waiting for an expressed practical interest. Neither C for loops nor repeat loops are supported.
Only build the total preconditions in a module without any interprocedural propagation. No specific condition must be met when reaching a RETURN statement.
Option total_preconditions_inter 6.9.3.3 uses the module own total postcondition derived from its callers as final state value and propagates it backwards in the module statement. This total module postcondition must be true when the RETURN statement is reached.
The program postcondition is only used for the main module.
The summary total precondition of a module is the total precondition of its statement limited to information observable by callers, just like a summary transformer (see Section 6.9.1.7).
A summary total precondition is of type ”transformer”.
A final postcondition for a module is derived from all its call sites. Of course, total postconditions must be known for all its callers’ statements. The summary total postcondition is the convex hull of all call sites total postconditions, translated into a proper environment which is not necessarily the module’s frame. Because of invisible global and static variables and aliasing, it is difficult for a caller to know which variables might be used by the caller to represent a given memory location. To avoid the problem, the current summary total postcondition is always translated into the caller’s frame. So each module must first translate its summary total postcondition, when receiving it from the resource manager (pipsdbm) before using it.
A summary total postcondition is of type ”transformer”.
The program postcondition cannot be derived from the source code. It should be defined explicitly by the user. By default, the predicate is always true. But you might want some variables to have specific values, e.g. KMAX==1, or signs,KMAX>1 or relationships KMAX>JMAX.
By default, the semantic analysis is restricted to scalar integer variables as they are key variables to understand scientific code behavior. However it is possible to analyze scalar variables with other data types. Fortran LOGICAL variables are represented as 0/1 integers. Character string constants and floating point constants are represented as undefined values.
The analysis is thus limited to constant propagation for character strings and floating point values whereas integer and boolean variables are processed with a relational analysis.
Character string constants of fixed maximal length could be translated into integers but the benefit is not yet assessed because they are not much used in the benchmark and commercial applications we have studied. The risk is to increase significantly the number of overflows encountered during the analysis.
SEMANTICS_ANALYZE_SCALAR_INTEGER_VARIABLES TRUE
SEMANTICS_ANALYZE_SCALAR_BOOLEAN_VARIABLES FALSE
SEMANTICS_ANALYZE_SCALAR_STRING_VARIABLES FALSE
SEMANTICS_ANALYZE_SCALAR_FLOAT_VARIABLES FALSE
SEMANTICS_ANALYZE_SCALAR_COMPLEX_VARIABLES FALSE
For every module, array declaration are assumed to be correct with respect to the standard: the upper bound must be greater than or equal to the lower bound. When implicit, the lower bound is one. The star upper bound is neglected.
This property is turned off by default because it might slow down PIPS quite a lot without adding any useful information because loop bounds are usually different from array bounds.
SEMANTICS_TRUST_ARRAY_DECLARATIONS FALSE
For every module, array references are assumed to be correct with respect to the declarations: the subscript expressions must have values lower than or equal to the upper bound and greater than or equal to the lower bound.
This property is turned off by default because it might slow down PIPS quite a lot without adding any useful information.
SEMANTICS_TRUST_ARRAY_REFERENCES FALSE
Type range information is difficult to turn into useful information. It implies some handling of wrap-around behavior. It is likely to cause lots of overflows. This is experimental. By default this property is not set.
This property is turned off by default because it might slow down PIPS quite a lot without adding any useful information.
SEMANTICS_USE_TYPE_INFORMATION FALSE
Perform “meet” operations for semantics analysis. This property is managed by pipsmake which often sets it to TRUE. See comments in pipsmake documentation to turn off convex hull operations for a module or more if they last too long.
SEMANTICS_FLOW_SENSITIVE FALSE
Complex control flow graph may require excessive computation resources. This may happen when analyzing a parser for instance.
SEMANTICS_ANALYZE_UNSTRUCTURED TRUE
To reduce execution time, this property is complemented with a heuristics to turn off the analysis of very complex unstructured.
If the control flow graph counts more than SEMANTICS_MAX_CFG_SIZE1 6.9.4.4 vertices, use effects only.
SEMANTICS_MAX_CFG_SIZE2 20
If the control flow graph counts more than SEMANTICS_MAX_CFG_SIZE1 6.9.4.4 but less than SEMANTICS_MAX_CFG_SIZE2 6.9.4.4 vertices, perform the convex hull of its elementary transformers and take the fixpoint of it. Note that SEMANTICS_MAX_CFG_SIZE2 6.9.4.4 is assumed to be greater than or equal to SEMANTICS_MAX_CFG_SIZE1 6.9.4.4.
SEMANTICS_MAX_CFG_SIZE1 20
Without preconditions, transformers can be precise only for affine expressions. Approximate transformers can sometimes be derived for other expressions, involving for instance products of variables or divisions.
However, a precondition of an expression can be used to refine the approximation. For instance, some non-linear expressions can become affine because some of the variables have constant values, and some non-linear expressions can be better approximated because the variables signs or ranges are known.
To be backward compatible and to be conservative for PIPS execution time, the default value is false.
Not implemented yet.
SEMANTICS_RECOMPUTE_EXPRESSION_TRANSFORMERS FALSE
Intraprocedural preconditions can be computed at the same time as transformers and used to improve the accuracy of expression and statement transformers. Non-linear expressions can sometimes have linear approximations over the subset of all possible stores defined by a precondition. In the same way, the number of convex hulls can be reduced if a test branch is never used or if a loop is always entered.
SEMANTICS_COMPUTE_TRANSFORMERS_IN_CONTEXT FALSE
The default value is false for reverse compatibility and for speed.
To be refined later; basically, use callee’s transformers instead of callee’s effects when computing transformers bottom-up in the call graph; when going top-down with preconditions, should we care about unique call site and/or perform meet operation on call site preconditions ?
SEMANTICS_INTERPROCEDURAL FALSE
This property is used internally and is not user selectable.
CPU time and memory space are cheap enough to compute loop fix points for transformers. This property implies SEMANTICS_FLOW_SENSITIVE 6.9.4.4 and is not user-selectable.
SEMANTICS_FIX_POINT FALSE
The default fix point operator, called transfer, is good for induction variables but it is not good for all kinds of code. The default fix point operator is based on the transition function associated to a loop body. A computation of eigenvectors for eigenvalue 1 is used to detect loop invariants. This fails when no transition function but only a transition relation is available. Only equations can be found.
The second fix point operator, called pattern, is based on a pattern matching of elementary equations and inequalities of the loop body transformer. Obvious invariants are detected. This fix point operator is not better than the previous one for induction variables but it can detect invariant equations and inequalities.
A third fix point operator, called derivative, is based on finite differences. It was developed to handled DO loops desugared into WHILE loops as well as standard DO loops. The loop body transformer on variable values is projected onto their finite differences. Invariants, both equations and inequalities, are deduced directly from the constraints on the differences and after integration. This third fix point operator should be able to find at least as many invariants as the two previous one, but at least some inequalities are missed because of the technique used. For instance, constraints on a flip-flop variable can be missed. Unlike Cousot-Halbwachs fix point (see below), it does not use Chernikova steps and it should not slow down analyses.
This property is user selectable and its default value is derivative. The default value is the only one which is now seriously maintained.
SEMANTICS_FIX_POINT_OPERATOR "derivative"
The next property is experimental and its default value is 1. It is used to unroll while loops virtually, i.e. at the semantics equation level, to cope with periodic behaviors such as flip-flops. It is effective only for standard while loops and the only possible value other than 1 is 2.
SEMANTICS_K_FIX_POINT 1
The next property SEMANTICS_PATTERN_MATCHING_FIX_POINT has been removed and replaced by option pattern of the previous property.
This property was defined to select one of Cousot-Halbwachs’s heuristics and to compute fix points with inequalities and equalities for loops. These heuristics could be used to compute fix points for transformers and/or preconditions. This option implies SEMANTICS_FIX_POINT 6.9.4.7 and SEMANTICS_FLOW_SENSITIVE 6.9.4.4. It has not been implemented yet in PIPS12 because its accuracy has not yet been required, but is now badly named because there is no direct link between inequality and Halbwachs. Its default value is false and it is not user selectable.
SEMANTICS_INEQUALITY_INVARIANT FALSE
Because of convexity, some fix points may be improved by using some of the information carried by the preconditions. Hence, it may be profitable to recompute loop fix point transformer when preconditions are being computed.
The default value is false because this option slows down PIPS and does not seem to add much useful information in general.
SEMANTICS_RECOMPUTE_FIX_POINTS_WITH_PRECONDITIONS FALSE
The next property is used to refine the computation of preconditions inside nested loops. The loop body is reanalyzed to get one transformer for each control path and the identity transformer is left aside because it is useless to compute the loop body precondition. This development is experimental and turned off by default.
SEMANTICS_USE_TRANSFORMER_LISTS FALSE
The next property is only useful if the previous one is set to true. Instead of computing the fix point of the convex hull of the transformer list, it computes the convex hull of the derivative constraints. Since it is a new feature, it is set to false by default, but it should become the default option because it should always be more accurate, at least indirectly because the systems are smaller. The number of overflows is reduced, as well as the execution time. In practice, these improvements have not been measured. This development is experimental and turned off by default.
SEMANTICS_USE_DERIVATIVE_LIST FALSE
The next property is only useful if Property SEMANTICS_USE_TRANSFORMER_LISTS 6.9.4.7 is set to true. Instead of computing the precondition derived from the transitive closure of a transformer list, semantics also computes the preconditions associated to different projections of the transformer list and use as loop precondition the intersection of these preconditions. Although it is a new feature, it is set to true by default for the validation’s sake. See test case Semantics/maisonneuve09.c: it improves the accuracy, but not as much as SEMANTICS_USE_DERIVATIVE_LIST 6.9.4.7. This development is experimental and turned off by default.
SEMANTICS_USE_LIST_PROJECTION TRUE
The string Property SEMANTICS_LIST_FIX_POINT_OPERATOR 6.9.4.7 is used to select a particular heuristic to compute an approximation of the transitive closure of a list of transformers. It is only useful if Property SEMANTICS_USE_TRANSFORMER_LISTS 6.9.4.7 is selected. The current default value is “depth_two”. An experimental value is “max_depth”.
SEMANTICS_LIST_FIX_POINT_OPERATOR "depth_two"
Preconditions can (used to) preserve initial values of the formal parameters. This is not often useful in C because programmers usually avoid modifying scalar parameters, especially integer ones. However, old values create problems in region computation because preconditions seem to be used instead of tranformer ranges. Filtering out the initial value does reduce the precision of the precondition analysis, but this does not impact the transformer analysis. Since the advantage is really limited to preconditions and for the region’s sake, the default value is set to true. Turn it to false if you have a doubt about the preconditions really available.
The loop index is usually dead on loop exit. So keeping information about its value is useless... most of the times. However, it is preserved by default.
SEMANTICS_KEEP_DO_LOOP_EXIT_CONDITION TRUE
SEMANTICS_FILTER_INITIAL_VALUES TRUE
Normalizing transformer and preconditions systems is a delicate issue which is not mathematically defined, and as such is highly empirical. It’s a tradeoff between eliminating redundant information, keeping an internal storage not too far from the prettyprinting for non-regression testing, exposing useful information for subsequent analyses,... all this at a reasonable cost.
Several levels of normalization are possible. These levels do not correspond to graduations on a normalization scale, but are different normalization heuristics. A level of 4 includes a preliminary lexicographic sort of contraints, which is very user friendly, but currently implies strings manipulations which are quite costly. It has been recently chosen to perform this normalization only before storing transformers and preconditions to the database (SEMANTICS_NORMALIZATION_LEVEL_BEFORE_STORAGE with a default value of 4). However, this can still have a serious impact on performances. With any other value, the normalization level is equel to 2.
SEMANTICS_NORMALIZATION_LEVEL_BEFORE_STORAGE 4
Preconditions reflect by default all knowledge gathered about the current state (i.e. store). However, it is possible to restrict the information to variables actually read or written, directly or indirectly, by the statement following the precondition.
SEMANTICS_FILTERED_PRECONDITIONS FALSE
Output semantics results on stdout
SEMANTICS_STDOUT FALSE
Debug level for semantics used to be controlled by a property. A Shell variable, SEMANTICS_DEBUG_LEVEL, is used instead.
Continuation conditions are attached to each statement. They represent the conditions under which the program will not stop in this statement. Under- and over-approximations of these conditions are computed.
Complexities are symbolic approximations of the execution times of statements. They are computed interprocedurally and based on polynomial approximations of execution times. Non-polynomial execution times are represented by unknown variables which are not free with respect to the program variables. Thus non-polynomial expressions are equivalent to polynomial expressions over a larger set of variables.
Probabilities for tests should also result in unknown variables (still to be implemented). See [?].
A summary_complexity is the approximation of a module execution times. It is translated and used at call sites.
Complexity estimation could be refined (i.e. the number of unknown variables reduced) by using transformers to combine elementary complexities using local states, rather than preconditions to combine elementary complexities relatively to the module initial state. The same options exist for region computation. The initial version [?] used the initial state for combinations. The new version [?] delays evaluation of variable values as long as possible but does not really use local states.
The first version of the complexity estimator was designed and developed by Pierre Berthomier. It was restricted to intra-procedural analysis. This first version was enlarged and validated on real code for SPARC-2 machines by Lei Zhou [?]. Since, it has been modified slightly by Franï¿œois Irigoin. For simple programs, complexity estimation are strongly correlated with execution times. The estimations can be used to see if program transformations are beneficial.
Known bugs: tests and while loops are not correctly handled because a fixed probably of 0.5 is systematically assumed.
Complexity estimation is based on a set of basic operations and fixed execution times for these basic operation. The choice of the set is critical but fixed. Experiments by Lei Zhou showed that it should be enlarged. However, the basic times, which also are critical, are tabulated. New sets of tables can easily be developed for new processors.
Uniform complexity tables contain a unit execution time for all basic operations. They nevertheless give interesting estimations for SPARC SS-10, especially for -O2/-O3 optimized code.
Local variables are eliminated from the complexity associated to the top statement of a module in order to obtain the modules’ summary complexity.
Tables for floating point complexity estimation are set to 0 for non-floating point operations, and to 1 for all floating point operations, including intrinsics like SIN.
This enables the default specification within the properties to be considered.
The following properties control the static estimation of dynamic code execution time.
Trace the walk across a module’s internal representation:
COMPLEXITY_TRACE_CALLS FALSE
Trace all intermediate complexities:
COMPLEXITY_INTERMEDIATES FALSE
Print the complete cost table at the beginning of the execution:
COMPLEXITY_PRINT_COST_TABLE FALSE
The cost table(s) contain machine and compiler dependent information about basic execution times, e.g. time for a load or a store.
It is possible to specify a list of variables which must remain literally in the complexity formula, although their numerical values are known (this is OK) or although they have multiple unknown and unrelated values during any execution (this leads to an incorrect result).
Formal parameters and imported global variables are left unevaluated.
They have relatively high priority (FI: I do not understand this comment by Lei).
This list should be empty by default (but is not for unknown historical reasons):
COMPLEXITY_PARAMETERS "IMAX␣LOOP"
Controls the printing of accuracy statistics:
COMPLEXITY_PRINT_STATISTICS 0
This property is used to select a set of basic execution times. These times depend on the target machine, the compiler and the compilation options used. It is shown in [?] that fixed basic times can be used to obtain accurate execution times, if enough basic times are considered, and if the target machine has a simple RISC processor. For instance, it is not possible to use only one time for a register load. It is necessary to take into account the nature of the variable, i.e. formal parameter, dynamic variable, global variable, and the nature of the access, e.g. the dimension of an accessed array. The cache can be ignored an replacer by an average hit ratio.
Different set of elementary cost tables are available:
In the future, we might add a sparc-2 table...
The different elementary table names are defined in complexity-local.h. They presently are operation, memory, index, transcend and trigo.
The different tables required are to be found in $PIPS_LIBDIR/complexity/xyz, where xyz is specified by this property:
COMPLEXITY_COST_TABLE "all_1"
For the moment, we have designed two ways to solve the complexity combination problem. Since symbolic complexity formulae use program variables it is necessary to specify in which store they are evaluated. If two complexity formulae are computed relatively to two different stores, they cannot be directly added.
The first approach, which is implemented, uses the module initial store as universal store for all formulae (but possibly for the complexity of elementary statements). In some way, symbolic variable are evaluated as early as possible as soon as it is known that they won’t make it in the module summary complexity.
This first method is easy to implement when the preconditions are available but it has at least two drawbacks:
The second approach, which is not implemented, delay variable evaluation as late as possible. Complexities are computed and given relatively to the stores used by each statements. Two elementary complexities are combined together using the earliest store. The two stores are related by a transformer (see Section 6.9.4). Such an approach is used to compute MUST regions as precisely as possible (see Section 6.12.9).
A simplified version of the late evaluation was implemented. The initial store of the procedure is the only reference store used as with the early evaluation, but variables are not evaluated right away. They only are evaluated when it is necessary to do so. This not an ideal solution, but it is easy to implement and reduces considerably the number of unknown values which have to be put in the formulae to have correct results.
COMPLEXITY_EARLY_EVALUATION FALSE
Convex array regions are functions mapping a memory store onto a convex set of array elements. They are used to represent the memory effects of modules or statements. Hence, they are expressed with respect to the initial store of the module or to the store immediately preceding the execution of the statement they are associated with. The latter is now standard in PIPS. Comprehensive information about convex array regions and their associated algorithms is available in Creusillet’s PhD Dissertation [?].
Apart from the array name and its dimension descriptors (or ϕ variables), an array region contains three additional informations:
Unfortunately, for historical reasons, MUST is still used in the implementation instead of EXACT, and actual MUST regions are not computed. Moreover, the must_regions option in fact computes exact and may regions.
MAY regions are flow-insensitive regions, whereas MUST regions are flow sensitive. Any array element touched by any execution of a statement is in the MAY region of this statement. Any array element in the MUST region of a statement is accessed by any execution of this statement.
For instance, the convex array region:
<A(ϕ1,ϕ2)-W-EXACT-{ϕ1==I, ϕ1==ϕ2}>
Internally, convex array regions are of type effect and as such can be used to build use-def chains (see Section 6.5.3). Regions chains are built using proper regions which are particular READ and WRITE regions. For simple statements (assignments, calls to intrinsic functions), summarization is avoided to preserve accuracy. At this inner level of the program control flow graph, the extra amount of memory necessary to store regions without computing their convex hull should not be too high compared to the expected gain for dependence analysis. For tests and loops, proper regions contain the regions associated to the condition or the range. And for external calls, proper regions are the summary regions of the callee translated into the caller’s name space, to which are merely appended the regions of the expressions passed as argument (no summarization for this step).
Together with READ/WRITE regions and IN regions are computed their invariant versions for loop bodies (MODULE.inv_regions and MODULE.inv_in_regions). For a given loop body, they are equal to the corresponding regions in which all variables that may be modified by the loop body (except the current loop index) are eliminated from the descriptors (convex polyhedron). For other statements, they are equal to the empty list of regions.
In the following trivial example,
notice that the variable k which is modified in the loop body, and which appears in the loop body region polyhedron, does not appear anymore in the invariant region polyhedron.
MAY READ and WRITE region analysis was first designed by Rï¿œmi Triolet [?] and then revisited by Franï¿œois Irigoin [?]. Alexis Platonoff [?] implemented the first version of region analysis in PIPS. These regions were computed with respect to the initial stores of the modules. Franï¿œois Irigoin and, mainly, Bï¿œatrice Creusillet [?, ?, ?], added new functionalities to this first version as well as functions to compute MUST regions, and IN and OUT regions.
Array regions for C programs are currently under development.
This function computes the MAY pointer regions in a module.
This function computes the MAY regions in a module.
This function computes the MUST regions in a module.
This function computes the MUST pointer regions in a module using simple points_to information to disambiguate dereferencing paths.
This function computes the MUST regions in a module.
This function computes the MUST regions in a module using information on pointer targets given by points-to.
This function computes the MUST regions in a module using information on pointer targets given by pointer values.
Module summary regions provides an approximation of the effects it’s execution has on its callers variables as well as on global and static variables of its callees.
IN convex array regions are flow sensitive regions. They are read regions not covered (i.e. not previously written) by assignments in the local hierarchical control-flow graph. There is no way with the current pipsmake-rc and pipsmake to express the fact that IN (and OUT) regions must be calculated using must_regions 6.12.3 (a new kind of resources, must_regions 6.12.3, should be added). The user must be knowledgeable enough to select must_regions 6.12.3 first.
This pass computes the IN convex array regions of a module. They contain the array elements and scalars whose values impact the output of the module.
This pass computes the OUT convex array regions of a module. They contain the array elements and scalars whose values impact the continuation of the module.
See Section 6.12.8.
OUT convex array regions are also flow sensitive regions. They are downward exposed written regions which are also used (i.e. imported) in the continuation of the program. They are also called exported regions. Unlike READ, WRITE and IN regions, they are propagated downward in the call graph and in the hierarchical control flow graphs of the subroutines.
If MUST_REGIONS is true, then it computes convex array regions using the algorithm described in report E/181/CRI, called T-1 algorithm. It provides more accurate regions, and preserve MUST approximations more often. As it is more costly, its default value is FALSE. EXACT_REGIONS is true for the moment for backward compatibility only.
EXACT_REGIONS TRUE
MUST_REGIONS FALSE
The default option is to compute regions without taking into account array bounds. The next property can be turned to TRUE to systematically add them in the region descriptors. Both options have their advantages and drawbacks.
REGIONS_WITH_ARRAY_BOUNDS FALSE
Property MEMORY_IN_OUT_EFFECTS_ONLY 6.12.9’s default value is set to TRUE to avoid computing IN and OUT effects or regions on non-memory effects, even if MEMORY_EFFECTS_ONLY 6.2.7 is set to FALSE.
MEMORY_IN_OUT_EFFECTS_ONLY TRUE
The current implementation of effects, simple effects as well as convex array regions, relies on a generic engine which is independent of the effect descriptor representation. The current representation for array regions, parameterized integer convex polyhedra, allows various patterns an provides the ability to exploit context information at a reasonable expense. However, some very common patterns such as nine-point stencils used in seismic computations or red-black patterns cannot be represented. It has been a long lasting temptation to try other representations [?].
A Complementary sections (see Section 6.14) implementation was formerly began as a set of new phases by Manjunathaiah Muniyappa, but is not maintained anymore.
And Nga Nguyen more recently created two properties to switch between regions and disjunctions of regions (she has already prepared basic operators). For the moment, they are always FALSE.
DISJUNCT_REGIONS FALSE
DISJUNCT_IN_OUT_REGIONS FALSE
Statistics may be obtained about the computation of convex array regions. When the next property (REGIONS_OP_STATISTICS) is set to TRUE statistics are provided about operators on regions (union, intersection, projection,…). The second next property turns on the collection of statistics about the interprocedural translation.
REGIONS_OP_STATISTICS FALSE
REGIONS_TRANSLATION_STATISTICS FALSE
Dynamic aliases are pairs (formal parameter, actual parameter) of convex array regions generated at call sites. An “IN alias pair” is generated for each IN region of a called module and an “OUT alias pair” for each OUT region. For EXACT regions, the transitive, symmetric and reflexive closure of the dynamic alias relation results in the creation of equivalence classes of regions (for MAY regions, the closure is different and does not result in an equivalence relation, but nonetheless allows us to define alias classes). A set of alias classes is generated for a module, based on the IN and OUT alias pairs of all the modules below it in the callgraph. The alias classes for the whole workspace are those of the module which is at the root of the callgraph, if the callgraph has a unique root. As an intermediate phase between the creation of the IN and OUT alias pairs and the creation of the alias classes, “alias lists” are created for each module. An alias list for a module is the transitive closure of the alias pairs (IN or OUT) for a particular path through the callgraph subtree rooted in this module.
This phase generates synthetic points-to relations for formal parameters. It creates synthetic sinks, i.e. stubs, for formal parameters and provides an initial set of points-to to the intraprocedural_points_to_analysis 6.13.3.
Currently, it assumes that no sharing exists between the formal parameters and within the data structures pointed to by the formal parameters. Two properties should control this behavior, ALIASING_ACROSS_FORMAL_PARAMETERS 6.13.7.1 and ALIASING_ACROSS_TYPES 6.13.7.1. The first one supersedes the property ALIASING_INSIDE_DATA_STRUCTURE 6.13.7.1.
This function is being implemented by Amira Mensi. The intraprocedural_points_to_analysis 6.13.3 is implemented in order to compute points-to relations, based on Emami algorithm. Emami algorithm is a top-down analysis which calcules the points-to relations by applying specific rules to each assignement pattern identified. This phase requires another resource which is init_points_to_analysis 6.13.2. Ressources points_to_in and points_to_out will be used to compute the transfer function later. They represent points-to relation at the beginning of functions where sources are formal parameters or global variables. Points_to_out are points-to relations at the end of function’s body, it contains return value and it’s sink, formal parameters, gloabl variables and heap allocated variables which can be visible beyond function’s scope. And using effects to compute calls impact on points-to analysis.
The pointer effects are useful, but they are recomputed for each expression and subexpression by the points-to analysis.
This pass is being implemented by Amira Mensi. The interprocedural_points_to_analysis 6.13.4 is implemented in order to compute points-to relations in an interprocedural way, based on Wilson algorithm. This phase computes both Gen and Kill sets at the level of the call site.It requires another resource which is computed by intraprocedural_points_to_analysis 6.13.3.
This pass is being implemented by Amira Mensi. The fast_interprocedural_points_to_analysis 6.13.5 is implemented in order to compute points-to relations in an interprocedural way, based on Wilson algorithm. This phase computes only Kill sets at the call site level. It requires another resource which is computed by intraprocedural_points_to_analysis 6.13.3.
Computes the initial pointer values from the global or static declarations of the module.
Computes the initial pointer values of the program from the global declarations and the static declarations inside the program modules. They are computed by merging the initial pointer values of all the modules (this may include those which do not belong to actually realizable paths).
Pointer values analysis is another kind of pointer analysis which tries to gather Pointer Values both in terms of other pointer values but also of memory addresses. This phase is under development.
The following properties are defined to ensure the safe use of intraprocedural_points_to_analysis 6.13.3.
The property ALIASING_ACROSS_TYPES 6.13.7.1 specifies that two pointers of different effective types can be aliased. The default and safe value is TRUE; when it is turned to FALSE two pointers of different types are never aliased.
ALIASING_ACROSS_TYPES TRUE
The property ALIASING_ACROSS_FORMAL_PARAMETERS 6.13.7.1 is used to handle the aliasing between formal parameters and global variables of pointer type. When it is set to TRUE, two formal parameters or a formal one and a global pointer or two global pointers can be aliased. If it is turned to FALSE, such pointers are assumed to be unaliased for intraprocedural analysis and generally for root module(i.e. modules without callers). The default value is FALSE. It is the only value currently implemented.
ALIASING_ACROSS_FORMAL_PARAMETERS FALSE
The nest property specifies that one data structure can recursively contain two pointers pointing to the same location. If it is turned to FALSE, it is assumed that two different not included memory access paths cannot point to the same memory locations. The safe value is TRUE, but parallelization is hindered. Often, the user can guarantee that data structures do not exhibit any sharing. Optimistically, FALSE is the default value.
ALIASING_INSIDE_DATA_STRUCTURE FALSE
Property ALIASING_ACROSS_IO_STREAMS 6.13.7.1 can be set to FALSE to specify that two io streams (two variables declared as FILE *) cannot be aliased, neither the locations to which they point. The safe and default value is TRUE
ALIASING_ACROSS_IO_STREAMS TRUE
The following string property defines the lattice of maximal elements to use when precise information is lost. Three values are possible: ”unique”, ”function” and ”area”. The first value is the default value. A unique identifier is defined to represent any set of unknown locations. The second value defines a separate identifier for each function and compilation unit. Note that compilation units require more explanation about this definition and about the conflict detection scheme. The third value, ”area”, requires a separate identifier for each area of each function or compilation unit. These abstract lcoation lattice values are further refined if the property ALIASING_ACROSS_TYPES 6.13.7.1 is set to FALSE. The abstract location API hides all these local maximal values from its callers. Note that the dereferencing of any such top abstract location returns the very top of all abstract locations.
The ABSTRACT_HEAP_LOCATIONS 6.13.7.2 specifies the modeling of the heap. The possible values are ”unique”, ”insensitive”, ”flow-sensitive” and ”context-sensitive”. Each value defines a stricly refined analysis with respect to analyses defined by previous values [This may not be a good idea, since flow and context sensitivity are orthogonal].
The default value, ”unique”, implies that the heap is a unique array. It is enough to parallelize simple loops containing pointer-based references such as ”p[i]”.
In the ”insensitive” case and all other cases, one array is allocated in each function to modelize the heap.
In the ”flow-sensitive” case, the statement numbers of the malloc() call sites are used to subscribe this array, as well as all indices of the surrounding loops [Two improvements in one property...].
In the ”context_sensitive” case, the interprocedural translation of memory acces paths based on the abstract heap are prefixed by the same information regarding the call site: function containing the call site, statement number of the call site and indices of surrounding loops.
Note that the naming of options is not fully compatible with the usual notations in pointer analyses. Note also that the insensitive case is redundant with context sensitive case: in the later case, a unique heap associated to malloc() would carry exactly the same amount of information [flow and context sensitivity are orthogonal].
Finally, note that abstract heap arrays are distinguished according to their types if the property ALIASING_ACROSS_TYPES 6.13.7.1 is set to FALSE [impact on abstract heap location API]. Else, the heap array is of type unknown. If a heap abstract location is dereferenced without any point-to information nor heap aliasing information, the safe result is the top abstract location.
ABSTRACT_HEAP_LOCATIONS "unique"
The property POINTS_TO_STRICT_POINTER_TYPES 6.13.7.2 is used to handle pointer arithmetic. According to C standard(section 6.5.6, item 8) the following C code :
is correct and p points to the same area, expressed by the points to analysis as i[*]. The default value is FALSE, meaning that p points to an array element. When it’s set to TRUE typing becone strict ; meaning that p points to an integer and the behavior is undefined. So the analysis stops with a pips_user_error(illegal pointer arithmetic)
POINTS_TO_STRICT_POINTER_TYPES FALSE
The property POINTS_TO_UNINITIALIZED_POINTER_DEREFERENCING 6.13.7.3 specifies what to do when an uninitialized pointer is or may be dereferenced. The safe value is FALSE. The points-to analysis assumed that no undefined pointer is ever dereferenced. So if a pointer may be undefined and is dereferenced, the arc is considered impossible and removed from the points-to information. If not other arc provides some value for this pointer, the code is assumed dead and the current points-to set is reduced to the empty set. A warning about dead code is emitted. However the property can be set to TRUE and the dereferencing of an undefined pointer is accepted and results in an anywhere location.
POINTS_TO_UNINITIALIZED_POINTER_DEREFERENCING FALSE
The property POINTS_TO_NULL_POINTER_DEREFERENCING 6.13.7.3 is very similar to the previous one. It specifies what to do when an null pointer is or may be dereferenced. The safe value is FALSE. The points-to analysis assumed that no null pointer is ever dereferenced. So if a pointer may be undefined and is dereferenced, the arc is considered impossible and removed from the points-to information. If not other arc provides some value for this pointer, the code is assumed dead and the current points-to set is reduced to the empty set. A warning about dead code is emitted. However the property can be set to TRUE and the dereferencing of an undefined pointer is accepted and results in an anywhere location.
POINTS_TO_NULL_POINTER_DEREFERENCING FALSE
The property POINTS_TO_NULL_POINTER_INITIALIZATION 6.13.7.3 allows the initialization of pointers that are formal parameters or global variables to NULL when computing a calling context. The most accurate property value is TRUE, which makes sure that generated points-to stubs are different from NULL because two arcs are always generated: an arc towards the new points-to stub and an arc towards the NULL location. Thus it prevents from dereferencing a null pointer when dereferencing a points-to stub and it allows the comparison of two points-to stubs when a condition such as p!=q is interpreted or the comparison of one points-to stub to NULL as in p!=NULL. Temporarily, to perform an interprocedural analysis, it must be turned to FALSE.
POINTS_TO_NULL_POINTER_INITIALIZATION TRUE
The integer property POINTS_TO_PATH_LIMIT 6.13.7.4 specifies the maximum number of occurences of an object of a given type in a non-cyclic path generated by the points-to graph. New nodes are created as long as no such path exists. When the limit is reached, a cycle is created.
POINTS_TO_PATH_LIMIT 2
The integer property POINTS_TO_SUBSCRIPT_LIMIT 6.13.7.4 specifies the maximum number of subscript of an object can be generated via pointer arithmetic. When the limit is reached, an unbounded subscript, *, is used to model any possible subscript value.
POINTS_TO_SUBSCRIPT_LIMIT 2
Display the dynamic alias pairs (formal region, actual region) for the IN regions of the module.
Display the dynamic alias pairs (formal region, actual region) for the OUT regions of the module.
Display the transitive closure of the dynamic aliases for the module.
Display the dynamic alias equivalence classes for this module and those below it in the callgraph.
A new representation of array regions added in PIPS by Manjunathaiah Muniyappa. This anlysis is not maintained anymore.
This function computes the complementary sections in a module.
Dynamic analyses are performed at run-time. At compile-time, the property than can be proved or disproved are exploited, but in doubt a run-time is added to the source code. The current dynamic analyses implemented in PIPSare array bound checking, Fortran alias and used-before-set analyses.
Array bound checking refers to determining whether all array references are within their declared range in all of their uses in a program. These array bound checks may be analysed intraprocedurally or interprocedurally, depending on the need for accuracy.
There are two versions of intraprocedural array bounds checking: array bound check bottom up, array bound check top down. The first approach relies on checking every array access and on the elimination of redundant tests by advanced dead code elimination based on preconditions. The second approach is based on exact convex array regions. They are used to prove that all accessed in a compound statement are correct.
These two dynamic analyses are implemented for Fortran. They are described in Nga Nguyen’s PhD (see [?]) and in [?]. They may work for C code, but this has not been validated.
This transformation takes as input the current module, adds array range checks (lower and upper bound checks) to every statement that has one or more array accesses. The output is the module with those added tests.
If one test is trivial or exists already for the same statement, it is no need to be generated in order to reduce the number of tests. As Fortran language permits an assumed-size array declarator with the unbounded upper bound of the last dimension, no range check is generated for this case also.
Associated with each test is a bound violation error message and in case of real access violation, a STOP statement will be put before the current statement.
This phase should always be followed by the partial_redundancy_elimination 9.2.2 for logical expression in order to reduce the number of bound checks.
This second implementation is based on the array region analyses phase which benefits some interesting proven properties:
If none of these two properties are satisfied, we consider the approximation of region. In case of MUST region, if the exact bound checks can be generated, they will be inserted before the block of code. If not, like in case of MAY region, we continue to go down to the children nodes in the control flow graph.
The main advantage of this algorithm is that it permits to detect the sure bound violations or to tell that there is certainly no bound violation as soon as possible, thanks to the context given by preconditions and the top-down analyses.
This phase checks for out of bound errors when passing arrays or array elements as arguments in procedure call. It ensures that there is no bound violation in every array access in the callee procedure, with respect to the array declarations in the caller procedure.
We provide here a tool to calculate the number of dynamic bound checks from both initial and PIPS generated code.
These transformations are implemented by Thi Viet Nga Nguyen (see [?]).
Array bounds checking refers to determining whether all array reference are within their declared range in all of its uses in a program. Here are array bounds checking options for code instrumentation, in order to compute the number of bound checks added. We can use only one property for these two case, but the meaning is not clear. To be changed ?
INITIAL_CODE_ARRAY_BOUND_CHECK_INSTRUMENTATION TRUE
PIPS_CODE_ARRAY_BOUND_CHECK_INSTRUMENTATION FALSE
In practice, bound violations may often occur with arrays in a common block. The standard is violated, but programmers think that they are not dangerous because the allocated size of the common is not reached. The following property deals with this kind of bad programming practice. If the array is a common variable, it checks if the reference goes beyond the size of the common block or not.
ARRAY_BOUND_CHECKING_WITH_ALLOCATION_SIZE FALSE
The following property tells the verification phases (array bound checking, alias checking or uninitialized variables checking) to instrument codes with the STOP or the PRINT message. Logically, if a standard violation is detected, the program will stop immediately. Furthermore, the STOP message gives the partial redundancy elimination phase more information to remove redundant tests occurred after this STOP. However, for the debugging purposes, one may need to display all possible violations such as out-of-bound or used-before-set errors, but not to stop the program. In this case, a PRINT message is chosen. By default, we use the STOP message.
PROGRAM_VERIFICATION_WITH_PRINT_MESSAGE FALSE
Aliasing occurs when two or more variables refer to the same storage location at the same program point. Alias analysis is critical for performing most optimizations correctly because we must know for certain that we have to take into account all the ways a location, or the value of a variable, may (or must) be used or changed. Compile-time alias information is also important for program verification, debugging and understanding.
In Fortran 77, parameters are passed by address in such a way that, as long as the actual argument is associated with a named storage location, the called subprogram can change the value of the actual argument by assigning a value to the corresponding formal parameter. So new aliases can be created between formal parameters if the same actual argument is passed to two or more formal parameters, or between formal parameters and global parameters if an actual argument is an object in common storage which is also visible in the called subprogram or other subprograms in the call chain below it.
Both intraprocedural and interprocedural alias determinations are important for program analysis. Intraprocedural aliases occur due to pointers in languages like LISP, C, C++ or Fortran 90, union construct in C or EQUIVALENCE in Fortran. Interprocedural aliases are generally created by parameter passing and by access to global variables, which propagates intraprocedural aliases across procedures and introduces new aliases.
The basic idea for computing interprocedural aliases is to follow all the possible chains of argument-parameters and nonlocal variable-parameter bindings at all call sites. We introduce a naming memory locations technique which guarantees the correctness and enhances the precision of data-flow analysis. The technique associates sections, offsets of actual parameters to formal parameters following a certain call path. Precise alias information are computed for both scalar and array variables. The analysis is called alias propagation.
This analysis is implemented by Thi Viet Nga Nguyen (see [?]).
With the call-by-reference mechanism in Fortran 77, new aliases can be created between formal parameters if the same actual argument is passed to two or more formal parameters, or between formal parameters and global parameters if an actual argument is an object in common storage which is also visible in the called subprogram or other subprograms in the call chain below it.
Restrictions on association of entities in Fortran 77 (Section 15.9.3.6 [?]) say that neither aliased formal parameters nor the variable in the common block may become defined during execution of the called subprogram or the others subprograms in the call chain.
This phase uses information from the alias_propagation 7.2.1 analysis and computes the definition informations of variables in a program, and then to verify statically if the program violates the standard restriction on alias or not. If these informations are not known at compile-time, we instrument the code with tests that check the violation dynamically during execution of program.
This verification is implemented by Thi Viet Nga Nguyen (see [?]).
This is a property to control whether the alias propagation and alias checking phases use information from MAIN program or not. If the current module is never called by the main program, we do no alias propagation and alias checking for this module if the property is on. However, we can do nothing with modules that have no callers at all, because this is a top-down approach.
ALIAS_CHECKING_USING_MAIN_PROGRAM FALSE
This analysis checks if the program uses a variable or an array element which has not been assigned a value. In this case, anything may happen: the program may appear to run normally, or may crash, or may behave unpredictably. We use IN regions that give a set of read variables not previously written. Depending on the nature of the variable: local, formal or global, we have different cases. In principle, it works as follows: if we have a MUST IN region at the module statement, the corresponding variable must be used before being defined, a STOP is inserted. Else, we insert an initialization function and go down, insert a verification function before each MUST IN at each sub-statements.
This is a top-down analysis that process a procedure before all its callees. Information given by callers is used to verify if we have to check for the formal parameters in the current module or not. In addition, we produce information in the resource MODULE.ubs to tell if the formal parameters of the called procedures have to be checked or not.
This verification is implemented by Thi Viet Nga Nguyen (see [?]).
PIPS basic parallelization function, rice_all_dependence 8.1.3, produces a new version of the Module code with DOALL loops exhibited using Allen & Kennedy’s algorithm. The DOALL syntactic construct is non-standard but easy to understand and usual in text book like [?]. As parallel prettyprinter option, it is possible to use Fortran 90 array syntax (see Section 10.4). For C, the loops can be output as for-loop decorated with OpenMP pragma.
Remember that Allen & Kennedy’s algorithm can only be applied on loops with simple bodies, i.e. sequences of assignments, because it performs loop distribution and loop regeneration without taking control dependencies into account. If the loop body contains tests and branches, the coarse grain parallelization algorithm should be used (see 8.1.6).
Loop index variables are privatized whenever possible, using a simple algorithm. Dependence arcs related to the index variable and stemming from the loop body must end up inside the loop body. Else, the loop index is not privatized because its final value is likely to be needed after the loop end and because no copy-out scheme is supported.
A better privatization algorithm for all scalar variable may be used as a preliminary code transformation. An array privatizer is also available (see Section 9.7.11). A non-standard PRIVATE declaration is used to specify which variables should be allocated on stack for each loop iteration. An HPF or OpenMP format can also be selected.
Objects of type parallelized_code differs from objects of type code for historic reasons, to simplify the user interface and because most algorithms cannot be applied on DOALL loops. This used to be true for pre-condition computation, dependence testing and so on... It is possible neither to re-analyze parallel code, nor to re-parse it (although it would be interesting to compute the complexity of a parallel code) right now but it should evolves. See § 8.1.8.
There are few properties that control the parallelization behaviour.
TRUE to make all possible parallel loops, FALSE to generate real (vector, innermost parallel?) code:
GENERATE_NESTED_PARALLEL_LOOPS TRUE
Show statistics on the number of loops parallelized by pips:
PARALLELIZATION_STATISTICS FALSE
To select whether parallelization and loop distribution is done again for already parallel loops:
PARALLELIZE_AGAIN_PARALLEL_CODE FALSE
The motivation is we may want to parallelize with a coarse grain method first, and finish with a fine grain method here to try to parallelize what has not been parallelized. When applying ￜ la Rice parallelizing to parallelize some (still) sequential code, we may not want loop distribution on already parallel code to preserve cache resources, etc.
Thread-safe libraries are protected by critical sections. Their functions can be called safely from different execution threads. For instance, a loop whose body contains calls to malloc can be parallelized. The underlying state changes do no hinder parallelization, at least if the code is not sensitive to pointer values.
PARALLELIZATION_IGNORE_THREAD_SAFE_VARIABLES FALSE
Since this property is used to mask arcs in the dependence graph, it must be exploited by each parallelization phase independently. It is not used to derived a simplified version of the use-def chains or of the dependence graph to avoid wrong result with use-def elimination, which is based on the same graph.
Entries in menu for the resource parallelized_code and for the different parallelization algorithms with may be activated or selected. Note that the nest parallelization algorithm is not debugged.
Use Allen & Kennedy’s algorithm and consider all dependences.
Several other parallelization functions for shared-memory target machines are available. Function rice_data_dependence 8.1.4 only takes into account data flow dependences, a.k.a true dependences. It is of limited interest because transitive dependences are computed. It is not equivalent at all to performing array and scalar expansion based on direct dependence computation (Brandes, Feautrier, Pugh). It is not safe when privatization is performed before parallelization.
This phase is named after the historical classification of data dependencies in output dependence, anti-dependence and true or data dependence. It should not be used for standard parallelization, but only for experimental parallelization by knowledgeable users, aware that the output code may be illegal.
Function rice_cray 8.1.5 targets Cray vector multiprocessors. It selects one outermost parallel loop to use multiple processors and one innermost loop for the vector units. It uses Cray microtasking directives. Note that a prettyprinter option must also be selected independently (see Section 10.4).
Function coarse_grain_parallelization 8.1.6 implements a loop parallelization algorithm based on convex array regions. It considers only one loop at a time, its body being abstracted by its invariant read and write regions. No loop distribution is performed, but any kind of loop body is acceptable whereas Allen & Kennedy algorithm only copes with very simple loop bodies.
For nasty reasons about effects that are statement addresses to effects mapping, this pass changes the code instead of producing a parallelized_code resource. It is not a big deal since often we want to modify the code again and we should use internalize_parallel_code 8.1.8 just after if its behavior were modified.
Function coarse_grain_parallelization_with_reduction 8.1.6 extend the standard coarse_grain_parallelization 8.1.6 by using reduction detection informations.
Function nest_parallelization 8.1.7 is an attempt at combining loop transformations and parallelization for perfectly nested loops. Different parameters are computed like loop ranges and contiguous directions for references. Loops with small ranges are fully unrolled. Loops with large ranges are strip-mined to obtain vector and parallel loops. Loops with medium ranges simply are parallelized. Loops with unknown range also are simply parallelized.
For each loop direction, the amount of spatial and temporal localities is estimated. The loop with maximal locality is chosen as innermost loop.
This algorithm still is in the development stage. It can be tried to check that loops are interchanged when locality can be improvedInternship!. An alternative for static control section, is to use the interface with PoCC (see Section 10.11).
To simplify the user interface and to display with one click a parallelized program, programs in PIPS are parallelized code instead of standard code.PV:not clear As a consequence, parallelized programs cannot be further analyzed and transformed because sequential code and parallelized code do not have the same resource type. Most pipsmake rules apply to code but not to parallelized code. Unfortunately, improving the parallelized code with some other transformations such as dead-code elimination is also useful. Thus this pseudo-transformation is added to coerce a parallel code into a classical (sequential) one. Parallelization is made an internal code transformation in PIPS with this rule.
Although this is not the effective process, parallel loops are tagged as parallel and loop local variables may be added in a code resource because of a previous privatization phase.
If you display the “generated” code, it may not be displayed as a parallel one if the PRETTYPRINT_SEQUENTIAL_STYLE 10.2.22.3.2 is set to a parallel output style (such as omp). Anyway, the information is available in code.
Note this transformation may no be usable with some special parallelizations in PIPS such as WP65 or HPFC that generate other resource types that may be quite different.
Generate a pragma on each loop that seems to be computation intensive according to a simple cost model.
The computation intensity is derived from the complexity and the memory footprint. It assumes the cost model:
This correspond to the transfer startup overhead. Time unit is the same as in complexities.
COMPUTATION_INTENSITY_STARTUP_OVERHEAD 10
This corresponds to the memory bandwidth in octet per time unit.
COMPUTATION_INTENSITY_BANDWIDTH 100
And This is the processor frequency, in operation per time unit.
COMPUTATION_INTENSITY_FREQUENCY 1000
This is the generated pragma.
COMPUTATION_INTENSITY_PRAGMA "pips␣intensive␣loop"
Those values have limited meaning here, only their ratio have some. Having COMPUTATION_INTENSITY_FREQUENCY 8.1.9 and COMPUTATION_INTENSITY_BANDWIDTH 8.1.9 of the same magnitude clearly limits the number of generated pragmas…
Parallel loops which are considered as not complex enough are replaced by sequential ones using a simple cost model based on complexity (see uniform_complexities 6.11.2).
This phase restricts the parallelism of parallel do-loop nests by limiting the number of top-level parallel do-loops to be below a given limit. The too many innermost parallel loops are replaced by sequential loops, if any. This is useful to keep enough coarse-grain parallelism and respecting some hardware or optimization constraints. For example on GPU, in CUDA there is a 2D limitation on grids of thread blocks, in OpenCL it is limited to 3D. Of course, since the phase works onto parallel loop nest, it might be interesting to use a parallelizing phase such as internalize_parallel_code (see § 8.1.8) or coarse grain parallelization before applying limit_nested_parallelism.
PIPS relies on the property NESTED_PARALLELISM_THRESHOLD 8.1.11 to determine the desired level of nested parallelism.
NESTED_PARALLELISM_THRESHOLD 0
The SAC project aims at generating efficient code for processors with SIMD extension instruction set such as VMX, SSE4, etc. which are also refered to as Superword Level Parallelism (SLP). For more information, see https://info.enstb.org/projets/sac, or better, see Serge Guelton’s PhD dissertation.
Some phases use ACCEL_LOAD 8.2 and ACCEL_STORE 8.2 to generate DMA calls and ACCEL_WORK 8.2.
ACCEL_LOAD "SIMD_LOAD"
ACCEL_STORE "SIMD_STORE"
ACCEL_WORK "SIMD_"
Here is yet another atomizer, based on new_atomizer (see Section 9.4.1.2), used to reduce complex statements to three-address code close to assembly code. There are only some minor differences with respect to new_atomizer, except that it does not break down simple expressions, that is, expressions that are the sum of a reference and a constant such as tt i+1. This is needed to generate code that could potentially be efficient, whereas the original atomizer would most of the time generate inefficient code.
Use the SIMD_ATOMIZER_ATOMIZE_REFERENCE 8.2.1 property to make the SIMD Atomizer go wild: unlike other atomizer, it will break the content of a reference. SIMD_ATOMIZER_ATOMIZE_LHS 8.2.1 can be used to tell the atomizer to atomize both lhs and rhs.
SIMD_ATOMIZER_ATOMIZE_REFERENCE FALSE
SIMD_ATOMIZER_ATOMIZE_LHS FALSE
The SIMD_OVERRIDE_CONSTANT_TYPE_INFERENCE 8.2.1 property is used by the SAC library to know if it must override C constant type inference. In C, an integer constant always as the minimum size needed to hold its value, starting from an int. In sac we may want to have it converted to a smaller size, in situation like char b;/*...*/;char a = 2 + b;. Otherwise the result of 2+b is considered as an int. if SIMD_OVERRIDE_CONSTANT_TYPE_INFERENCE 8.2.1 is set to TRUE, the result of 2+b will be a char.
SIMD_OVERRIDE_CONSTANT_TYPE_INFERENCE FALSE
Tries to unroll the code for making the simdizing process more efficient. It thus tries to compute the optimal unroll factor, allowing to pack the most instructions together. Sensible to SIMDIZER_AUTO_UNROLL_MINIMIZE_UNROLL 8.2.11.1 and SIMDIZER_AUTO_UNROLL_SIMPLE_CALCULATION 8.2.11.1.
Similiar to simdizer_auto_unroll 8.2.2 but at the loop level.
Sensible to LOOP_LABEL 9.1.1.
Tries to tile the code to make the simdizing process more efficient.
Sensible to LOOP_LABEL 9.1.1 to select the loop nest to tile.
This phase tries to pre-process reductions, so that they can be vectorized efficiently by the simdizer 8.2.10 phase. When multiple reduction statements operating on the same variable with the same operation are detected inside a loop body, each “instance” of the reduction is renamed, and some code is added before and after the loop to initialize the new variables and compute the final result.
SIMD_REMOVE_REDUCTIONS_PREFIX "RED"
SIMD_REMOVE_REDUCTIONS_PRELUDE ""
SIMD_REMOVE_REDUCTIONS_POSTLUDE ""
Remove useless load store calls (and more).
If REDUNDANT_LOAD_STORE_ELIMINATION_CONSERVATIVE 8.2.5 is set to false, redundant_load_store_elimination 8.2.5 will remove any statement not implied in the computation of out regions, otherwise it will not remove statement that modifies aprameters reference.
REDUNDANT_LOAD_STORE_ELIMINATION_CONSERVATIVE TRUE
...
This phase is the first phase of the if-conversion algorithm. The complete if conversion algorithm is performed by applying the three following phase: if_conversion_init 8.2.7, if_conversion 8.2.7 and if_conversion_compact 8.2.7.
Use IF_CONVERSION_INIT_THRESHOLD 8.2.7 to control whether if conversion will occur or not: beyhond this number of call, no conversion is done.
IF_CONVERSION_INIT_THRESHOLD 40
This phase is the second phase of the if-conversion algorithm. The complete if conversion algorithm is performed by applying the three following phase: if_conversion_init 8.2.7, if_conversion 8.2.7 and if_conversion_compact 8.2.7.
IF_CONVERSION_PHI "__C-conditional__"
This phase is the third phase of the if-conversion algorithm. The complete if conversion algorithm is performed by applying the three following phase: if_conversion_init 8.2.7, if_conversion 8.2.7 and if_conversion_compact 8.2.7.
The Scalar Renaming pass tries to minimize dependencies in the code by renaming scalars when legal.
This function initialize a treematch used by simdizer 8.2.10 for simd-oriented pattern matching
Function simdizer 8.2.10 is an attempt at generating SIMD code for SIMD multimedia instruction set such as MMX, SSE2, VIS,... This transformation performs the core vectorization, transforming sequences of similar statements into vector operations.
When set to true, following property tells the simdizer to try to padd arrays when it seems to be profitable
SIMDIZER_ALLOW_PADDING FALSE
Skip generation of load and stores, using generic functions instead.
SIMDIZER_GENERATE_DATA_TRANSFERS TRUE
This phase is to be called after simdization of affectation operator. It performs type substitution from char/short array to in array using the packing from the simdization phase For example, four consecutive load from a char array could be a single load from an int array. This prove to be useful for c to vhdl compilers such as c2h.
This property is used to set the target register size, expressed in bits, for places where this is needed (for instance, auto-unroll with simple algorithm).
SAC_SIMD_REGISTER_WIDTH 64
This property is used to control how the auto unroll phase computes the unroll factor. By default, the minimum unroll factor is used. It is computed by using the minimum of the optimal factor for each statement. If the property is set to FALSE, then the maximum unroll factor is used instead.
SIMDIZER_AUTO_UNROLL_MINIMIZE_UNROLL TRUE
This property controls how the “optimal” unroll factor is computed. Two algorithms can be used. By default, a simple algorithm is used, which simply compares the actual size of the variables used to the size of the registers to find out the best unroll factor. If the property is set to FALSE, a more complex algorithm is used, which takes into account the actual SIMD instructions.
SIMDIZER_AUTO_UNROLL_SIMPLE_CALCULATION TRUE
This property is used by the sac library to know which elements of multi-dimensional array are consecutive in memory. Let us consider the three following references a(i,j,k), a(i,j,k+1) and a(i+1,j,k). Then, if SIMD_FORTRAN_MEM_ORGANISATION 8.2.11.2 is set to TRUE, it means that a(i,j,k) and a(i+1,j,k) are consecutive in memory but a(i,j,k) and a(i,j,k+1) are not. However, if SIMD_FORTRAN_MEM_ORGANISATION 8.2.11.2 is set to FALSE, a(i,j,k) and a(i,j,k+1) are consecutive in memory but a(i,j,k) and a(i+1,j,k) are not.
SIMD_FORTRAN_MEM_ORGANISATION TRUE
This property is used by the sac library to know the path of the pattern definition file. If the file is not found, the execution fails.
SIMD_PATTERN_FILE "patterns.def"
Different automatic code distribution techniques are implemented in PIPS for distributed-memory machines. The first one is based on the emulation of a shared-memory. The second one is based on HPF. A third one target architectures with hardware coprocessors. Another one is currently developed at IT Sud Paris that generate MPI code from OpenMP one.
WP651 [?, ?, ?] produces a new version of a module transformed to be executed on a distributed memory machine. Each module is transformed into two modules. One module, wp65_compute_file, performs the computations, while the other one, wp65_bank_file, emulates a shared memory.
This rule does not have data structure outputs, as the two new program generated have computed names. This does not fit the pipsmake framework too well, but is OK as long as nobody wishes to apply PIPS on the generated code, e.g. to propagate constant or eliminate dead code.
Note that use-use dependencies are used to allocate temporary arrays in local memory (i.e. in the software cache).
This compilation scheme was designed by Corinne Ancourt and Franï¿œois Irigoin. It uses theoretical results in [?]. Its input is a very small subset of Fortran program (e.g. procedure calls are not supported). It was implemented by the designers, with help from Lei Zhou.
Name of the file for the target model:
WP65_MODEL_FILE "model.rc"
The HPF compiler2 is a project by itself, developed by Fabien Coelho in the PIPS framework.
A whole set of rules is used by the PIPS HPF compiler3 , HPFC4 . By the way, the whole compiler is just a big hack according to Fabien Coelho.
The first rule is used to apply a shell to put HPF-directives in an f77 parsable form. Some shell script based on sed is used. The hpfc_parser 4.2.2 must be called to analyze the right file. This is triggered automatically by the bang selection in the hpfc_close 8.3.2.5 phase.
The second HPFC rule is used to initialize the hpfc status and other data structures global to the compiler. The HPF compiler status is bootstrapped. The compiler status stores (or should store) all relevant information about the HPF part of the program (data distribution, IO functions and so on).
This phase removes the directives (some special calls) from the code. The remappings (implicit or explicit) are also managed at this level, through copies between differently shaped arrays.
To manage calls with distributed arguments, I need to apply the directive extraction bottom-up, so that the callers will know about the callees through the hpfc_status. In order to do that, I first thought of an intermediate resource, but there was obscure problem with my fake calls. Thus the dependence static then dynamic directive analyses is enforced at the bang sequence request level in the hpfc_close 8.3.2.5 phase.
The hpfc_static_directives 8.3.2.3 phase analyses static mapping directives for the specified module. The hpfc_dynamic_directives 8.3.2.3 phase does manages realigns and function calls with prescriptive argument mappings. In order to do so it needs its callees’ required mappings, hence the need to analyze beforehand static directives. The code is cleaned from the hpfc_filter 8.3.2.1 artifacts after this phase, and all the proper information about the HPF stuff included in the routines is stored in hpfc_status.
This rule launches the actual compilation. Four files are generated:
Between this phase and the previous one, many PIPS standard analyses are performed, especially the regions and preconditions. Then this phase will perform the actual translation of the program into a host and SPMD node code.
This rule deals with the compiler closing. It must deal with commons. The hpfc parser selection is put here.
This rule performs the installation of HPFC generated files in a separate directory. This rule is added to make hpfc usable from wpips and epips. I got problems with the make and run rules, because it was trying to recompute everything from scratch. To be investigated later on.
Debugging levels considered by HPFC: HPFC_{,DIRECTIVES,IO,REMAPPING}_DEBUG_LEVEL.
These booleans control whether some computations are directly generated in the output code, or computed through calls to dedicated runtime functions. The default is the direct expansion.
HPFC_EXPAND_COMPUTE_LOCAL_INDEX TRUE
HPFC_EXPAND_COMPUTE_COMPUTER TRUE
HPFC_EXPAND_COMPUTE_OWNER TRUE
HPFC_EXPAND_CMPLID TRUE
HPFC_NO_WARNING FALSE
Hacks control…
HPFC_FILTER_CALLEES FALSE
GLOBAL_EFFECTS_TRANSLATION TRUE
These booleans control the I/O generation.
HPFC_SYNCHRONIZE_IO FALSE
HPFC_IGNORE_MAY_IN_IO FALSE
Whether to use lazy or non-lazy communications
HPFC_LAZY_MESSAGES TRUE
Whether to ignore FCD (Fabien Coelho Directives…) or not. These directives are used to instrument the code for testing purposes.
HPFC_IGNORE_FCD_SYNCHRO FALSE
HPFC_IGNORE_FCD_TIME FALSE
HPFC_IGNORE_FCD_SET FALSE
Whether to measure and display the compilation times for remappings, and whether to generate outward redundant code for remappings. Also whether to generate code that keeps track dynamically of live mappings. Also whether not to send data to a twin (a processor that holds the very same data for a given array).
HPFC_TIME_REMAPPINGS FALSE
HPFC_REDUNDANT_SYSTEMS_FOR_REMAPS FALSE
HPFC_OPTIMIZE_REMAPPINGS TRUE
HPFC_DYNAMIC_LIVENESS TRUE
HPFC_GUARDED_TWINS TRUE
Whether to use the local buffer management. 1 MB of buffer is allocated.
HPFC_BUFFER_SIZE 1000000
HPFC_USE_BUFFERS TRUE
Whether to use in and out convex array regions for input/output compiling
HPFC_IGNORE_IN_OUT_REGIONS TRUE
Whether to extract more equalities from a system, if possible.
HPFC_EXTRACT_EQUALITIES TRUE
Whether to try to extract the underlying lattice when generating code for systems with equalities.
HPFC_EXTRACT_LATTICE TRUE
RK: IT SudParis : insert your documentation here; FI: or a pointer towards you documentation
The step_parser 8.3.3.1 phase identifies the OpenMP constructs. The directive semantics are stored in the MODULE.step_directives ressource.
The step_analyse_init 8.3.3.2 phase init the PROGRAM.step_comm ressources
The step_analyse 8.3.3.2 phase triggers the convex array regions analyses to compute SEND and RECV regions leading to MPI messages and checks whether a given SEND region corresponding to a directive construct is consumed by a RECV region corresponding to a directive construct. In this case, communications can be optimized.
Based on the OpenMP construct and analyses, new modules are generated to translate the original code with OpenMP directives. The default code transformation for OpenMP construct is driven by the STEP_DEFAULT_TRANSFORMATION 8.3.3.3 property. The different value allowed are :
STEP_DEFAULT_TRANSFORMATION "HYBRID"
The step_compile 8.3.3.3 phase generates source code for OpenMP constructs depending of the transformation desired. Each OpenMP construct could have a specific transformation define by STEP clauses (without specific clauses, the STEP_DEFAULT_TRANSFORMATION 8.3.3.3 is used). The specific STEP clauses allowed are :
The step_install 8.3.3.3 phase copy the generated source files in the directory specified by the STEP_INSTALL_PATH 8.3.3.3 property.
STEP_INSTALL_PATH ""
The PHRASE project is an attempt to automatically (or semi-automatically) transform high-level language programs into code with partial execution on some accelerators such as reconfigurable logic (such as FPGAs) or data-paths.
This phases allow to split the code into portions of code delimited by PHRASE-pragma (written by the programmer) and a control program managing them. Those portions of code are intended, after transformations, to be executed in reconfigurable logic. In the PHRASE project, the reconfigurable logic is synthesized with the Madeo tool that take SmallTalk code as input. This is why we have a SmallTalk pretty-printer (see section 10.10).
This phase is a preparation phase for the Phrase Distributor phrase_distributor 8.3.4.2: the portions of code to externalize are identified and isolated here. Comments are modified by this phase.
This phase is automatically called by the following phrase_distributor 8.3.4.2.
The job of distribution is done here. This phase should be applied after the initialization (Phrase Distributor Initialisation phrase_distributor_init 8.3.4.1), so this one is automatically applied first.
This phase add control code for PHRASE distribution. All calls to externalized code portions are transformed into START and WAIT calls. Parameters communication (send and receive) are also handled here
The Safescale project is an attempt to automatically (or semi-automatically) transform sequential code written in C language for the Kaapi runtime.
This phase is intended for the analysis of a module given with the aim of finding blocks of code delimited by specific pragmas from it.
This phase is intended for the externalization of a block of code.
This phase should be applied after the initialization (Phrase Distributor Initialisation or phrase_distributor_init 8.3.4.1). The job of comEngine distribution is done here.
This property is set to TRUE if we want to synthesize only one process on the HRE.
COMENGINE_CONTROL_IN_HRE TRUE
This property holds the fifo size of the ComEngine.
COMENGINE_SIZE_OF_FIFO 128
Isolate the statement given in ISOLATE_STATEMENT_LABEL 8.3.7.1 in a separated memory. Data transfer are generated using the same DMA as kernel_load_store 8.3.7.4.
The algorithm is based on Read and write regions (no in / out yet) to compute the data that must be copied and allocated. Rectangular hull of regions are used to match allocator and data transfers prototypes. If an analysis fails, definition regions are use instead. If a sizeof is involved, EVAL_SIZEOF 9.4.2 must be set to true.
ISOLATE_STATEMENT_LABEL ""
As a side effect of isolate_statement pass, some new variables are declared into the function. A prefix can be used for the names of those variables using the property ISOLATE_STATEMENT_VAR_PREFIX. It is also possible to insert a suffix using the property ISOLATE_STATEMENT_VAR_SUFFIX. The suffix will be inserted between the original variable name and the instance number of the copy.
ISOLATE_STATEMENT_VAR_PREFIX ""
ISOLATE_STATEMENT_VAR_SUFFIX ""
By default we cannot isolate a statement with some complex effects on the non local memory. But if we know we can (for example ), we can override this behaviour by setting the following property:
ISOLATE_STATEMENT_EVEN_NON_LOCAL FALSE
Optimize the load/store dma by delaying the stores and performing the stores as soon as possible. Interprocedural version.
It uses ACCEL_LOAD 8.2 and ACCEL_STORE 8.2 to distinguish loads and stores from other calls.
The communication elimination makes the assumption that a load/store pair can always be removed.
Optimize the load/store dma by delaying the stores and performing the stores as soon as possible. Intra Procedural version.
It uses ACCEL_LOAD 8.2 and ACCEL_STORE 8.2 to distinguish loads and stores from other calls.
The communication elimination makes the assumption that a load/store pair can always be removed.
if SOLVE_HARDWARE_CONSTRAINTS_TYPE 8.3.7.3 is set to VOLUME, Given a loop label, a maximum memory footprint and an unknown entity, try to find the best value for SOLVE_HARDWARE_CONSTRAINTS_UNKNOWN 8.3.7.3 to make memory footprint of SOLVE_HARDWARE_CONSTRAINTS_LABEL 8.3.7.3 reach but not exceed SOLVE_HARDWARE_CONSTRAINTS_LIMIT 8.3.7.3. If it is set to NB_PROC, it tries to find the best value for SOLVE_HARDWARE_CONSTRAINTS_UNKNOWN 8.3.7.3 to make the maximum range of first dimension of all regions accessed by SOLVE_HARDWARE_CONSTRAINTS_LABEL 8.3.7.3 equals to SOLVE_HARDWARE_CONSTRAINTS_LIMIT 8.3.7.3.
SOLVE_HARDWARE_CONSTRAINTS_LABEL ""
SOLVE_HARDWARE_CONSTRAINTS_LIMIT 0
SOLVE_HARDWARE_CONSTRAINTS_UNKNOWN ""
SOLVE_HARDWARE_CONSTRAINTS_TYPE ""
Bootstraps the kernel ressource
Add a kernel to the list of kernels known to pips
Generate unoptimized load / store information for each call to the module.
The legacy kernel_load_store 8.3.7.4 approach is limited because it generates the DMA around a call, and isolate_statement 8.3.7.1 engine does not perform well in interprocedural.
The following properties are used to specify the names of runtime functions. Since they are used in Par4All, their default names begin with P4A_. To have an idea about their prototype, have a look to the Par4All accelerator runtime or in validation/AcceleratorUtils/include/par4all.c.
Enable/disable the scalar handling by kernel load store.
KERNEL_LOAD_STORE_SCALAR FALSE
The ISOLATE_STATEMENT_EVEN_NON_LOCAL 8.3.7.1 property can be used to force the generation even with non local memory access. But beware it would not solve all the issues...
The following properties can be used to customized the allocate/load/store functions:
KERNEL_LOAD_STORE_ALLOCATE_FUNCTION "P4A_accel_malloc"
KERNEL_LOAD_STORE_DEALLOCATE_FUNCTION "P4A_accel_free"
The following properties are used to name the dma functions to use for scalars:
KERNEL_LOAD_STORE_LOAD_FUNCTION "P4A_copy_to_accel"
KERNEL_LOAD_STORE_STORE_FUNCTION "P4A_copy_from_accel"
and for 1-dimension arrays:
KERNEL_LOAD_STORE_LOAD_FUNCTION_1D "P4A_copy_to_accel_1d"
KERNEL_LOAD_STORE_STORE_FUNCTION_1D "P4A_copy_from_accel_1d"
and in 2 dimensions:
KERNEL_LOAD_STORE_LOAD_FUNCTION_2D "P4A_copy_to_accel_2d"
KERNEL_LOAD_STORE_STORE_FUNCTION_2D "P4A_copy_from_accel_2d"
and in 3 dimensions:
KERNEL_LOAD_STORE_LOAD_FUNCTION_3D "P4A_copy_to_accel_3d"
KERNEL_LOAD_STORE_STORE_FUNCTION_3D "P4A_copy_from_accel_3d"
and in 4 dimensions:
KERNEL_LOAD_STORE_LOAD_FUNCTION_4D "P4A_copy_to_accel_4d"
KERNEL_LOAD_STORE_STORE_FUNCTION_4D "P4A_copy_from_accel_4d"
and in 5 dimensions:
KERNEL_LOAD_STORE_LOAD_FUNCTION_5D "P4A_copy_to_accel_5d"
KERNEL_LOAD_STORE_STORE_FUNCTION_5D "P4A_copy_from_accel_5d"
and in 6 dimensions:
KERNEL_LOAD_STORE_LOAD_FUNCTION_6D "P4A_copy_to_accel_6d"
KERNEL_LOAD_STORE_STORE_FUNCTION_6D "P4A_copy_from_accel_6d"
As a side effect of kernel load store pass, some new variables are declared into the function. A prefix can be used for the names of those variables using the property KERNEL_LOAD_STORE_VAR_PREFIX 8.3.7.4. It is also possible to insert a suffix using the property KERNEL_LOAD_STORE_VAR_PREFIX 8.3.7.4. The suffix will be inserted between the original variable name and the instance number of the copy.
KERNEL_LOAD_STORE_VAR_PREFIX "p4a_var_"
KERNEL_LOAD_STORE_VAR_SUFFIX ""
Split a parallel loop with a local index into three parts: a host side part, a kernel part and an intermediate part. The intermediate part simulates the parallel code to the kernel from the host
KERNELIZE_NBNODES 128
KERNELIZE_KERNEL_NAME ""
KERNELIZE_HOST_CALL_NAME ""
OUTLINE_LOOP_STATEMENT FALSE
Gather all constants from a module and put them in a single array. Relevant for Terapix code generation, and maybe for other accelerators as well
You may want to group constants only for a particular statement, in that case use GROUP_CONSTANTS_STATEMENT_LABEL 8.3.7.4
GROUP_CONSTANTS_STATEMENT_LABEL ""
The way variables are grouped is control by GROUP_CONSTANTS_LAYOUT 8.3.7.4, the only relevant value as of now is "terapix".
GROUP_CONSTANTS_LAYOUT ""
The name of the variable holding constants can be set using GROUP_CONSTANTS_HOLDER 8.3.7.4.
GROUP_CONSTANTS_HOLDER "caillou"
You may want to skip loop bounds from the grouping
GROUP_CONSTANTS_SKIP_LOOP_RANGE FALSE
You may want to skip litterals too.
GROUP_CONSTANTS_LITERAL TRUE
Perform various checks on a Terapix microcode to make sure it can be synthesized. GROUP_CONSTANTS_HOLDER 8.3.7.4 is used to differentiate mask and image.
This pass is meaningless for any other target :(.
converts divide operator into multiply operator using formula a∕cste = a * (1∕b) ≃ a * (128∕cste)∕128
TERAPIX_REMOVE_DIVIDE_ACCURACY 4
This phase computes the mapping of data on the accelarators. It records the set of data that have to be copied on the GPU before each statement in the module, and the set of data that have to be copied back from the GPU after the execution of each statement.
Then according to this information, the copy-in and copy-out transfers are generated using same set of properties as kernel_load_store 8.3.7.4.
This work has been described in [?][?]. However, the implementation is more complex than the published equations because of PIPS’ HCFG and because of a heuristic to generate transfers as high as possible in the HCFG.
This phase wrap argument at call site with an access function. The wrapper name is controlled with WRAP_KERNEL_ARGUMENT_FUNCTION_NAME 8.3.7.5. Currently the purpose of this is to insert call to a runtime to resolve addresses in accelerator memory corresponding to addresses in host memory.
WRAP_KERNEL_ARGUMENT_FUNCTION_NAME "P4A_runtime_host_ptr_to_accel_ptr"
This phase generate GPU kernels from perfect parallel loop nests. GPU_IFY_ANNOTATE_LOOP_NESTS 8.3.8 property triggers automatically the annotation of the loop nest (see gpu_loop_nest_annotate 8.3.8).
GPU_IFY_ANNOTATE_LOOP_NESTS FALSE
For example from
it generates something like
The launcher, wrapper and kernel prefix names to be used during the generation:
GPU_LAUNCHER_PREFIX "p4a_launcher"
GPU_WRAPPER_PREFIX "p4a_wrapper"
GPU_KERNEL_PREFIX "p4a_kernel"
This boolean property control wherever the outliner use the original function name as a suffix instead of only numerical suffix.
GPU_OUTLINE_SUFFIX_WITH_OWNER_NAME TRUE
For Fortran output you may need to have these prefix name in uppercase.
Indeed, each level of outlining can be enabled or disabled according to the following properties:
GPU_USE_LAUNCHER TRUE
GPU_USE_WRAPPER TRUE
GPU_USE_KERNEL TRUE
Each generated function can go in its own source file according to the following properties:
GPU_USE_KERNEL_INDEPENDENT_COMPILATION_UNIT FALSE
GPU_USE_LAUNCHER_INDEPENDENT_COMPILATION_UNIT FALSE
GPU_USE_WRAPPER_INDEPENDENT_COMPILATION_UNIT FALSE
By default they are set to FALSE for languages like CUDA that allow kernel and host codes mixed in a same file but for OpenCL it is not the case.
When the original code is in Fortran it might be useful to wrap the kernel launcher in an independent C file. The GPU_USE_FORTRAN_WRAPPER 8.3.8 can be used for that purpose. The name of the function wrapper can be configured using the property GPU_FORTRAN_WRAPPER_PREFIX 8.3.8. As specified before it is safe to use prefix name in uppercase.
GPU_USE_FORTRAN_WRAPPER FALSE
GPU_FORTRAN_WRAPPER_PREFIX "P4A_FORTRAN_WRAPPER"
The phase generates a wrapper function to get the iteration coordinate from intrinsics functions instead of the initial loop indices. Using this kind of wrapper is the normal behaviour but for simulation of an accelerator code, not using a wrapper is useful.
The intrinsics function names to get an ith coordinate in the iteration space are defined by this GNU ￜ la printf format:
GPU_COORDINATE_INTRINSICS_FORMAT "P4A_vp_%d"
where %d is used to get the dimension number. Here vp stands for virtual processor dimension and is a reminiscence from PompC and HyperC...
Please, do not use this feature for buffer-overflow attack...
Annotates loop nests with comments and guards for further generation of CUDA calls.
To annotate only outer parallel loop nests, set the following variable to true:
GPU_LOOP_NEST_ANNOTATE_PARALLEL TRUE
Clear annotation previously added by gpu_loop_nest_annotate 8.3.8.
Parallelize annotated loop nests based on the sentinel comments.
This phase promote sequential code in GPU kernels to avoid memory transfers.
The SCMP architecture, an asymmetric multiprocessor system-on-chip for dynamic applications, is described in [?]. SESAM is a simulation tool built up to help the design of such architectures. It relies on a specific programming model based on the explicit separation of the control and computation tasks, and its HAL provides high level memory allocation, shared memory access and synchronization functions.
The goal of the project was to generate applications for this architecture from sequential C code. Two different approaches were implemented. The first tries to identify tasks but is more specific. The second one is more general but tasks must have been previously identified with labels; this is currently performed manually.
The goal of the following phase is to generate SCMP tasks from C functions. The tasks are linked and scheduled using the SCMP Hardware Adaptation Layer (HAL). Pass sesamify 8.3.9.1 takes as input a module and analyzes all its callees. For instance, the ’main’ module can be submitted to sesamify after the gpu_ify 8.3.8 or scalopragma 8.3.9.1 pass have been applied. Each analyzed module is transformed into a SCMP task if its name begins with P4A_scmp_task. To generate the final files for the SCMP simulator, the pass output must be transformed by a specific python parser.
This pass outlines code parts based on pragma. It can outline blocs or loops with a #pragma scmp task flag. It is based on the outline pass.
The goal of Bufferization is to generate a dataflow communication through buffers between modules. The communication is done by special function call generated by kernel_load_store 8.3.7.4. To keep flows consistent outside the module scalopify 8.3.9.1 surrounds variable call with a special function too. A C file with stubs is needed.
Note that you must also set KERNEL_LOAD_STORE_DEALLOCATE_FUNCTION 8.3.7.4 to ”” in order to have it generate relevant code.
The goal of this pass is to keep consistent flows outside the tasks.
This code generation flow first relies on phase isolate_statement 8.3.7.1 to isolate the memory spaces of tasks identified by labels beginning by SCALOPES_KERNEL_TASK_PREFIX 8.3.9.2. Then phase sesam_buffers_processing 8.3.9.2 generates a header file describing how kernel and server tasks use the SESAM shared buffers. A post-processing phase is necessary to actually generate all the tasks of the distributed application. This is implemented in Par4All.
This solution is extensively described in [?]. A less technical presentation can be found in [?].
Phase sesam_buffers_processing is to be run after isolate_statement has been applied to all tasks statements. It then produces a header file to be included by the future SESAM application individual tasks. This header file describes how kernel and server tasks use the SESAM buffers.
The next two properties are used by phase sesam_buffers_processing to detect kernel tasks statements in the input module and to generate server tasks names in the output header file.
SCALOPES_KERNEL_TASK_PREFIX "P4A_sesam_task_"
SCALOPES_SERVER_TASK_PREFIX "P4A_sesam_server_"
A program transformation is a special phase which takes a code as input, modifies it, possibly using results from several different analyses, and puts back this modified code as result.
A rule describing a program transformation will never be chosen automatically by pipsmake to generate some code since every transformation rule contains a cycle for the MODULE.code resource. Since the first rule producing code, described in this file, is controlizer 4.3 and since it is the only non-cyclic rule, the internal representation always is initialized with it.
As program transformations produce nothing else, pipsmake cannot guess when to apply these rules automatically. This is exactly what the user want most of the time: program transformations are under explicit control by the user. Transformations are applied when the user pushes one of wpips transformation buttons or when (s)he enters an apply command when running tpips1 , or by executing a Perform Shell script. See the introduction for pointers to the user interfaces.
Unfortunately, it is sometime nice to be able to chain several transformations without any user interaction. No general macro mechanism is available in pipsmake, but it is possible to impose some program transformations with the ’!’ command.
User inputs are not well-integrated although a user_query rule and a string resource could easily be added. User interaction with a phase are performed directly without notifying pipsmake to be more flexible and to allow dialogues between a transformation and the user.
Most loop transformations require the user to give a valid loop label to locate the loop to be transformed. This is done interactively or by setting the following property to the valid label:
LOOP_LABEL ""
Put a label on unlabelled loops for further interactive processing. Unless FLAG_LOOPS_DO_LOOPS_ONLY 9.1.1 is set to false, only do loops are considered.
FLAG_LOOPS_DO_LOOPS_ONLY TRUE
Display label of all modules loops
Use intermediate variables as loop upper and lower bound when they are not affine.
Function distributer 9.1.3 is a restricted version of the parallelization function rice* (see Section 8.1.3).
Distribute all the loops of the module.
Allen & Kennedy’s algorithm [?] is used in both cases. The only difference is that distributer 9.1.3 does not produce DOALL loops, but just distributes loops as much as possible.
Partial distribution distributes the statements of a loop nest except the isolated statements,that have no dependences at the common level l, are gathered in the same l-th loop.
PARTIAL_DISTRIBUTION FALSE
Check if the statement flagged by STATEMENT_INSERTION_PRAGMA 9.1.4 can be safely inserted in the current control flow. This pass should be reserved to internal use only, another pass should create and insert a flagged statement and then call this one to verify the validity of the insertion
STATEMENT_INSERTION_PRAGMA "pips␣inserted␣statement␣to␣check"
STATEMENT_INSERTION_SUCCESS_PRAGMA "pips␣inserted␣statement"
STATEMENT_INSERTION_FAILURE_PRAGMA "pips␣inserted␣statement␣to␣remove"
Prepare the loop expansion by creating a new statement (that may be invalid) for further processing by statement_insertion 9.1.4. Use STATEMENT_INSERTION_PRAGMA 9.1.4 to identify the created statement. Otherwise LOOP_LABEL 9.1.1 and LOOP_EXPANSION_SIZE 9.1.5 have the same meaning as in loop_expansion 9.1.5
Extends the range of a loop given by LOOP_LABEL 9.1.1 to fit a size given by LOOP_EXPANSION_SIZE 9.1.5. An offset can be set if LOOP_EXPANSION_CENTER 9.1.5 is set to True. The new loop is guarded to prevent illegal iterations, further transformations can elaborate on this.
LOOP_EXPANSION_SIZE ""
LOOP_EXPANSION_CENTER FALSE
Extends the dimension of all declared arrays so that no access is illegal.
This pass fuses as many loops as possible in a greedy manner. The loops must appear in a sequence and have exactly the same loop bounds and if possible the same loop indices. We’ll always try first to fuse loops where there is a dependence between their body. We expect that this policy will maximize possibilities for further optimizations.
Property LOOP_FUSION_GREEDY 9.1.6 allows to control whether it’ll try to fuse as many loop as possible even without any reuse. This will be done in a second pass.
Property LOOP_FUSION_MAXIMIZE_PARALLELISM 9.1.6 is used to control if loop fusion has to preserve parallelism while fusing. If this property is true, a parallel loop is never fused with a sequential loop.
Property LOOP_FUSION_KEEP_PERFECT_PARALLEL_LOOP_NESTS 9.1.6 prevents to lose parallelism when fusing outer loops from a loop nests without being able to fuse inner loops.
Property LOOP_FUSION_MAX_FUSED_PER_LOOP 9.1.6 limit the number of fusion per loop. A negative value means that no limit will be enforced.
The fusion legality is checked in the standard way by comparing the dependence graphs obtained before and after fusion.
This pass is still in the experimental stage. It may have side effects on the source code when the fusion is attempted but not performed in case loop index are different.
This pass is the same is loop_fusion 9.1.6 excepts that it uses regions instead of dependence graph. Properties are the same as before: LOOP_FUSION_GREEDY 9.1.6, LOOP_FUSION_MAXIMIZE_PARALLELISM 9.1.6, LOOP_FUSION_KEEP_PERFECT_PARALLEL_LOOP_NESTS 9.1.6, and LOOP_FUSION_MAX_FUSED_PER_LOOP 9.1.6 control the algorithm.
LOOP_FUSION_MAXIMIZE_PARALLELISM TRUE
LOOP_FUSION_GREEDY FALSE
LOOP_FUSION_KEEP_PERFECT_PARALLEL_LOOP_NESTS TRUE
LOOP_FUSION_MAX_FUSED_PER_LOOP -1
Index Set Splitting [?] splits the loop referenced by property LOOP_LABEL 9.1.1 into two loops. The first loop ends at an iteration designated by property INDEX_SET_SPLITTING_BOUND 9.1.7 and the second start thereafter. It currently only works for do loops. This transformation is always legal. Index set splitting in combination with loop unrolling could be used to perform loop peeling.
Index Set Splitting requires the following globals to be set :
INDEX_SET_SPLITTING_BOUND ""
Additionnaly, INDEX_SET_SPLITTING_SPLIT_BEFORE_BOUND 9.1.7 can be used to accurately tell to split the loop before or after the bound given in INDEX_SET_SPLITTING_BOUND 9.1.7
INDEX_SET_SPLITTING_SPLIT_BEFORE_BOUND FALSE
Unroll requests a loop label and an unrolling factor from the user. Then it unrolls the specified loop as specified. The transformation is very general, and it is interesting to run partial_eval 9.4.2, simplify_control 9.3.1 and dead_code_elimination 9.3.2 after this transformation. When the number of iterations cannot be proven to be a multiple of the unrolling factor, the extra iterations can be executed first or last (see LOOP_UNROLL_WITH_PROLOGUE 9.1.8.1).
Labels in the body are deleted. To unroll nested loops, start with the innermost loop.
This transformation is always legal.
Use LOOP_LABEL 9.1.1 and UNROLL_RATE 9.1.8.1 if you do not want to unroll interactively You can also set LOOP_UNROLL_MERGE 9.1.8.1 to use the same declarations among all the unrolled statement (only meaningful in C).
UNROLL_RATE 0
LOOP_UNROLL_MERGE FALSE
The unrolling rate does not always divide exactly the number of iterations. So an extra loop must be added to execute the remaining iterations. This extra loop can be executed with the first iterations (prologue option) or the last iterations (epilogue option). Property LOOP_UNROLL_WITH_PROLOGUE 9.1.8.1 can be set to FALSE to use the epilogue when possible. The current implementation of the unrolling with prologue is general, while the implementation of the unrolling with epilogue is restricted to loops with a statically knonw increment of one. The epilogue option may reduce misalignments.
LOOP_UNROLL_WITH_PROLOGUE TRUE
Another option might be to require unrolling of the prologue or epilogue loop when possible.
A loop can also be fully unrolled if the range is numerically known. “Partial Eval” may be usefully applied first.
This is only useful for small loop ranges.
Unrolling can be interactively applied and the user is requested a loop label:
Or directives can be inserted as comments for loops to be unrolled with:
Full loop unrolling is applied one loop at a time by default. The user must specify the loop label. This default feature can be turned off and all loops with constant loop bounds and constant increment are fully unrolled.
Use LOOP_LABEL 9.1.1 to pass the desired label if you do not want to give it interactively
Property FULL_LOOP_UNROLL_EXCEPTIONS 9.1.8.2 is used to forbid loop unrolling when specific user functions are called in the loop body. The function names are separated by SPACEs. The default value is the empy set, i.e. the empry string.
FULL_LOOP_UNROLL_EXCEPTIONS ""
This pass applies unconditionnally a loop fusion between the loop designated by the property LOOP_LABEL 9.1.1 and the following loop. They must have the same loop index and the same iteration set. No legality check is performed.
Strip-mine requests a loop label and either a chunk size or a chunk number. Then it strip-mines the specified loop, if it is found. Note that the DO/ENDDO construct is not compatible with such local program transformations.
Behavior of strip mining can be controlled by the following properties:
STRIP_MINE_KIND -1
STRIP_MINE_FACTOR -1
loop_interchange 9.1.11 requests a loop label and exchange the outer-most loop with this label and the inner-most one in the same loop nest, if such a loop nest exists.
Presently, legality is not checked.
Property LOOP_LABEL 9.1.1 can be set to a loop label instead of using the default interactive method.
loop_hyperplane 9.1.12 requests a loop label and a hyperplane direction vector and applies the hyperplane method to the loop nest starting with this loop label, if such a loop nest exists.
Presently, legality is not checked.
loop_tiling 9.1.13 requests from the user a numerical loop label and a numerical partitioning matrix and applies the tiling method to the loop nest starting with this loop label, if such a loop nest exists.
The partitioning matrix must be of dimension n × n where n is the loop nest depth. The default origin for the tiling is 0, but lower loop bounds are used to adjust it and decrease the control overhead. For instance, if each loop is of the usual kind, DO I = 1, N, the tiling origin is point (1, 1,...). The code generation is performed according to the PPoPP’91 paper but redundancy elimination may results in different loop bounds.
Presently, legality is not checked. There is no decision procedure to select automatically an optimal partitioning matrix. Since the matrix must be numerically known, it is not possible to generate a block distribution unless all loop bounds are numerically known. It is assumed that the loop nest is fully parallel.
Jingling Xue published an advanced code generation algorithm for tiling in Parallel Processing Letters (http://cs.une.edu.au/~xue/pub.html).
This transformations prompts the user for a partition matrix. Alternatively, this matrix can be provided through the LOOP_TILING_MATRIX 9.1.13 property. The format of the matrix is a00 a01 a02,a10 a11 a12,a20 a21 a22
LOOP_TILING_MATRIX ""
Likewise, one can use the LOOP_LABEL 9.1.1 property to specify the targeted loop.
Tiles a loop nest using a partitioning vector that can contain symbolic values. The tiling only works for parallelepiped tiles. Use LOOP_LABEL 9.1.1 to specify the loop to tile. Use SYMBOLIC_TILING_VECTOR 9.1.14 as a comma-separated list to specify tile sizes. Use SYMBOLIC_TILING_FORCE 9.1.14 to bypass condition checks. Consider using loop_nest_unswitching 8.2.7 if generated max disturbs further analyses
SYMBOLIC_TILING_VECTOR ""
SYMBOLIC_TILING_FORCE FALSE
The loop normalization consists in transforming all the loops of a given module into a normal form. In this normal form, the lower bound and the increment are equal to one (1).
Property LOOP_NORMALIZE_PARALLEL_LOOPS_ONLY 9.1.15 control whether we want to normalize only parallel loops or all loops.
If we note the initial DO loop as:
| DO I = lower, upper, incre |
| ... |
| ENDDO |
the transformation gives the folowing code:
| DO | NLC = 0, (upper - lower + incre)/incre - 1, 1 |
| I = incre*NLC + lower |
| ... |
| ENDDO |
| I = | incre * MAX((upper - lower + incre)/incre, 0) + lower |
The normalization is done only if the initial increment is a constant number. The normalization produces two assignment statements on the initial loop index. The first one (at the beginning of the loop body) assigns it to its value function of the new index and the second one (after the end of the loop) assigns it to its final value.
If the increment is 1, the loop is considered already normalized. To have a 1-increment loop normalized too, set the following property
LOOP_NORMALIZE_ONE_INCREMENT FALSE
This is useful to have iteration spaces that begin at 0 for GPU for example.
The loop normalization has been defined in some days only Fortran was available, so having loops starting at 1 like the default for arrays too make sense in Fortran.
Anyway, no we could generalize for C (starting at 0 is more natural) or why not from any other value that can be chosen with the following property:
LOOP_NORMALIZE_LOWER_BOUND 1
If you are sure the final assignment is useless, you can skip it with the following property.
LOOP_NORMALIZE_SKIP_INDEX_SIDE_EFFECT FALSE
LOOP_NORMALIZE_PARALLEL_LOOPS_ONLY FALSE
Youcef Bouchebaba’s implementation of unimodular loop transformations…
Tiling for sequences of loop nests
Youcef Bouchebaba’s implementation of tiling for sequences of loop nests …
This is a test to implement a loop-invariant code motion. This phase hoist loop-invariant code out of the loop.
A side effect of this transformation is that the code is parallelized too with some loop distribution. If you don’t want this side effect, you can check section ?? which does a pretty nice job too.
The original algorithm used is described in Chapters 12, 13 and 14 of Julien Zory’s PhD dissertation [?].
Note: this pass deals with loop invariant code motion while the icm pass deals with expressions.
In essence, a partial redundancy [?] is a computation that is done more than once on some path through a flowgraph. We implement here a partial redundancy elimination transformation for logical expressions such as bound checks by using informations given by precondition analyses.
This transformation is implemented by Thi Viet Nga Nguyen.
See also the transformation in § sec:comm-subexpr-elim-1, the partial evaluation, and so on.
Function simplify_control 9.3.1 is used to delete non-executed code, such as empty loop nests or zero-trip loops, for example after strip-mining or partial evaluation.
Preconditions are used to find always true conditions in tests and to eliminate such tests. In some cases, tests cannot be eliminated, but test conditions can be simplified. One-trip loops are replaced by an index initialization and the loop body. Zero-trip loops are replaced by an index initialization. Effects in bound computations are preserved.
A lot of dead code can simply be eliminated by testing its precondition feasibility. A very simple and fast test may be used if the preconditions are normalized when they are computed, but this slows down the precondition computation. Or non-normalized preconditions are stored in the database and an accurate and slow feasibility test must be used. Currently, the first option is used for assignments, calls, IOs and IF statements but a stronger feasibility test is used for loops.
FORMAT statements are suppressed because they behave like a NOP command. They should be gathered at the beginning or at the end of the module using property GATHER_FORMATS_AT_BEGINNING 4.3 or GATHER_FORMATS_AT_END 4.3. The property must be set before the control flow graph of the module is computed.
The cumulated effects are used in debug mode to display information.
The simplify_control 9.3.1 phase also performs some If Simplifications and Loop Simplifications [?].
This function was designed and implemented by Ronan Keryell.
This pass is the same as simplify_control 9.3.1. It is used under this obsolete name in some validation scripts. The name has been preserved for backward compatibility.
This pass is very similar to simplify_control 9.3.1, but it does not require the preconditions. Only local information is used. It can be useful to clean up input code with constant tests, e.g. 3>4, and constant loop bounds. It can also be used after partial_eval 9.4.2 to avoid recomputing the preconditions yet another time. The property SIMPLIFY_CONTROL_DIRECTLY_PRIVATE_LOOP_INDICES 9.3.1 assert that the loop indices don’t need a copy out, i.e. the value at the exit of the loop can be forgotten.
SIMPLIFY_CONTROL_DIRECTLY_PRIVATE_LOOP_INDICES FALSE
Whether to try to simplify do/while loops (property introduced as a special workaround for FREIA).
SIMPLIFY_CONTROL_DO_WHILE TRUE
It is sometimes useful to display statistics on what has been found useless and removed in a function, this property controls the statistics display:
DEAD_CODE_DISPLAY_STATISTICS TRUE
Function dead_code_elimination 9.3.2 deletes statements whose def references are all dead, i.e. are not used by later executions of statements. It was developed by Ronan Keryell. The algorithm compute the set of live statements without fix-point. An initial set of live statements is extended with new statements reached thru use-def chains, control dependences and....
The initial set of live statements contains IO statements, RETURN, STOP,...
Note that use-def chains are computed intraproceduraly and not interproceduraly. Hence some statements may be preserved because they update a formal parameter although this formal parameter is no longer used by the callers.
The dependence graph may be used instead of the use-def chains, but Ronan Keryell, designer and implementer of the initial Fortran version, did not produce convincing evidence of the benefit... The drawback is the additional CPU time required.
This pass was extended to C by Mehdi Amini in 2009-2010, but it is not yet stabilized. For C code, this pass requires that effects are calculated with property MEMORY_EFFECTS_ONLY set to FALSE because we need that the DG includes arcs for declarations as these latter are separate statements now.
clean_declarations 9.7.1 is automatically done at the end, this is why cumulated effects are needed.
Comments from Nga Nguyen: According to [?] p. 595, and [?] p. 592, a variable is dead if it is not used on any path from the location in the code where it is defined to the exit point of the routine in the question; an instruction is dead if it computes only values that are not used on any executable path leading from the instruction. The transformation that identifies and removes such dead code is called dead code elimination. So in fact, the Use-def elimination pass in PIPS is a Dead code elimination pass and the Suppress dead code pass (see Section 9.3.1) does not have a standard name. It could be If and loop simplification pass.
The following properties are intended to force some functions to be kept by the algorithm, DEAD_CODE_ELIMINATION_KEEP_FUNCTIONS 9.3.2 expect a space separated list of function names while DEAD_CODE_ELIMINATION_KEEP_FUNCTIONS 9.3.2 expect a space separated list of prefix for function name.
DEAD_CODE_ELIMINATION_KEEP_FUNCTIONS ""
DEAD_CODE_ELIMINATION_KEEP_FUNCTIONS_PREFIX ""
For backward compatibility, the previous pass name is preserved.
Two control restructurers are available: unspaghettify 9.3.3.1 which is used by default in conjunction with controlizer 4.3 and restructure_control 9.3.3.2 which must be explicitly applied2
The unspaghettifier is a heuristic to clean up and to simplify the control graphs of a module. It is useful because the controlizer (see Section 4.3) or some transformation phases can generate some spaghetti code with a lot of useless unstructured code which can confuse some other parts of PIPS. Dead code elimination, for example, uses unspaghettify 9.3.3.1.
This control restructuring transformation can be automatically applied in the controlizer 4.3 phase (see Section 4.3) if the UNSPAGHETTIFY_IN_CONTROLIZER 4.3 property is true.
To add flexibility, the behavior of unspaghettify 9.3.3.1 is controlled by the properties UNSPAGHETTIFY_TEST_RESTRUCTURING 9.3.3.1 and UNSPAGHETTIFY_RECURSIVE_DECOMPOSITION 9.3.3.1 to allow more restructuring from restructure_control 9.3.3.2 to be added in the controlizer 4.3 for example.
This function was designed and implemented by Ronan Keryell.
To display the statistics about unspaghettify 9.3.3.1 and control graph restructuring restructure_control 9.3.3.2.
UNSPAGHETTIFY_DISPLAY_STATISTICS TRUE
The following option enables the use of IF/THEN/ELSE restructuring when applying unspaghettify:
UNSPAGHETTIFY_TEST_RESTRUCTURING FALSE
It is assumed as true for restructure_control 9.3.3.2. It recursively implement TEST restructuring (replacing IF/THEN/ELSE with GOTOs with structured IF/THEN/ELSE without any GOTOs when possible) by applying pattern matching methods.
The following option enables the use of control graph hierarchisation when applying unspaghettify:
UNSPAGHETTIFY_RECURSIVE_DECOMPOSITION FALSE
It is assumed as true for restructure_control 9.3.3.2. It implements a recursive decomposition of the control flow graph by an interval graph partitioning method.
The restructurer can recover some while loops if this property is set:
UNSPAGHETTIFY_WHILE_RECOVER FALSE
restructure_control 9.3.3.2 is a more complete restructuring phase that is useful to improve the accuracy of various PIPS phases.
It is implemented by calling unspaghettify 9.3.3.1 (§ 9.3.3.1) with the properties UNSPAGHETTIFY_TEST_RESTRUCTURING 9.3.3.1 and UNSPAGHETTIFY_RECURSIVE_DECOMPOSITION 9.3.3.1 set to TRUE.
Other restructuring methods are available in PIPS with the TOOLPACK’s restructurer (see Section 9.3.4).
This control-flow transformation transforms while loops into DO loops by recovering an index variable, an initial value, a final value and an increment.
Useful to be run after transformations ?!?
This phase cannot be called from inside the control restructurer since it needs many higher-level analysis. This is why it is in a separate phase.
Since in PIPS some transformations and analysis are more precise for Fortran code, this is a transformation than try to transform the C-like for-loops into Fortran-like do-loops.
Don’t worry about the C-code output: the prettyprinter output do-loop as for-loop if the C-output is selected. The do-loop construct is interesting since the iteration set is computed at the loop entry (for example it is not sensible to the index modification from the inside of the loop) and this simplifies abstract interpretation a lot.
This transformation transform for example a
into a
Since in PIPS some transformations and analysis may not be implemented for C for loops but may be implemented for while loops, it is interesting to have this for loop to while loop conversion.
This transformation transforms a
into a
Since analysis are more precise on do-loops, you should apply a for_loop_to_do_loop 9.3.3.4 transformation first , and only after, apply this for_loop_to_while_loop 9.3.3.5 transformation that will transform the remaining for-loops into while loops.
Some transformations only work on while loops, thus it is useful to have this transformation that transforms a
into a
It is a transformation useful before while loop to for loop recovery for example (see § 9.3.3.3).
spaghettify 9.3.3.7 is used in the context of the PHRASE project while creating “Finite State Machine”-like code portions in order to synthesize them in reconfigurable units.
This phases transform structured code portions (eg. loops) in unstructured statements.
spaghettify 9.3.3.7 transforms the module in a unstructured code with hierarchical unstructured portions of code corresponding to the old control flow structures.
To add flexibility, the behavior of spaghettify 9.3.3.7 is controlled by the properties
to allow more or less destruction power.
Thoses properties allow to fine tune spaghettify 9.3.3.7 phase
DESTRUCTURE_TESTS TRUE
DESTRUCTURE_LOOPS TRUE
DESTRUCTURE_WHILELOOPS TRUE
DESTRUCTURE_FORLOOPS TRUE
The spaghettify 9.3.3.7 is used in context of PHRASE project while creating“Finite State Machine”-like code portions in order to synthesize them in reconfigurable units.
This phases transforms all the module in a unique flat unstructured statement.
Whereas the spaghettify 9.3.3.7 transforms the module in a unstructured code with hierarchical unstructured portions of code corresponding to the old structures, the full_spaghettify 9.3.3.8 transform the code in a sequence statement with a beginning statement, a unique and flattened unstructured (all the unstructured and sequences are flattened), and a final statement.
This pass is now obsolete. Use restructure_control 9.3.3.2 instead.
Transformation stf 9.3.4 is a C interface to a Shell script used to restructure a Fortran program using ISTST (via the combined tool fragment ISTLY = ISTLX/ISTYP and then ISTST) from TOOLPACK [?, ?].
Be careful, since TOOLPACK is written in Fortran, you need the Fortran runtime libraries to run STF if is has not been statically compiled...
Known bug/feature: stf 9.3.4 does not change resource code like other transformations, but the source file. Transformations applied before stf 9.3.4 are lost.
This transformation is now assumed redundant with respect to the native PIPS control restructurers, which deal with other languages too.
Function suppress_trivial_test 9.3.5 is used to delete the TRUE branch of trivial test instruction. After apply suppress_trivial_test 9.3.5, the condition of the new test instruction is the condition correspondent to the FALSE branch of the initial test.
This function was designed and implemented by Trinh Quoc Anh.
Theses phases are used for PHRASE project.
NB: The PHRASE project is an attempt to automatically (or semi-automatically) transform high-level language for partial evaluation in reconfigurable logic (such as FPGAs or DataPaths).
This library provides phases allowing to build and modify ”Finite State Machine”-like code portions which will be later synthesized in reconfigurable units. This was implemented by Sylvain Guï¿œrin.
This phase tries to generate finite state machine from arbitrary code by applying rules numeroting branches of the syntax tree and using it as state variable for the finite state machine.
This phase recursively transforms each UNSTRUCTURED statement in a WHILE-LOOP statement controlled by a state variable, whose different values are associated to the different statements.
To add flexibility, the behavior of fsm_generation 9.3.6.1 is controlled by the property FSMIZE_WITH_GLOBAL_VARIABLE 9.3.6.5 which controls the fact that the same global variable (global to the current module) must be used for each FSMized statements.
To generate a hierarchical finite state machine, apply first spaghettify 9.3.3.7 (§ 9.3.3.7) and then fsm_generation 9.3.6.1.
To generate a flat finite state machine, apply first full_spaghettify 9.3.3.8 (§ 9.3.3.8) and then fsm_generation 9.3.6.1 or use the aggregate phase full_fsm_generation 9.3.6.2.
This phase tries to generate a flat finite state machine from arbitrary code by applying rules numeroting branches of the syntax tree and using it as state variable for the finite state machine.
This phase transform all the module in a FSM-like code, which is a WHILE-LOOP statement controlled by a state variable, whose different values are associated to the different statements.
In fact, this phase do nothing but rely on pipsmake to apply the succession of the 2 phases full_spaghettify 9.3.3.8 and fsm_generation 9.3.6.1 (§ 9.3.6.1)
This phase is not yet implemented and do nothing right now...
This phase transform a state of a FSM-like statement and split it into n new states where the portion of code to execute is smaller.
NB: Phase full_spaghettify 9.3.3.8 must have been applied first !
This phase is not yet implemented and do nothing right now...
This phase transform 2 or more states of a FSM-like statement and merge them into a new state where the portion of code to execute is bigger.
NB: Phase full_spaghettify 9.3.3.8 must have been applied first !
Control the fact that the same global variable (global to the current module) must be used for each FSMized statements.
FSMIZE_WITH_GLOBAL_VARIABLE FALSE
A code instrumentation that adds local integer counters in tests and loops to know how many times a path is taken. This transformation may help some semantical analyses.
Atomizer produces, or should produce, three-address like instructions, in Fortran. An atomic instructions is an instruction that contains no more than three variables, such as A = B op C. The result is a program in a low-level Fortran on which you are able to use all the others passes of PIPS.
Atomizers are used to simplify the statement encountered by automatic distribution phases. For instance, indirect addressing like A(B(I)) = ... is replaced by T=B(I);A(T) = ....
This pass performs subscripts atomization so that they can be converted in reference for more accruate analysis.
This pass evaluates expression of the form *”ae” that can be found in COLD output.
I doubt it can be useful elsewhere ...
This pass performs a conversion from complex to real. SIMPLIFY_COMPLEX_USE_ARRAY_OF_STRUCTS 9.4.1.2 controls the new layout
SIMPLIFY_COMPLEX_USE_ARRAY_OF_STRUCTS TRUE
Split structures in separated variables when possible, that is remove the structure variable and replaces all fields by different variables.
Here is a new version of the atomizer using a small atomizer from the HPF compiler (see Section 8.3.2).
An atomizer is also used by WP65 (see Section 8.3.1)
This transformation only atomizes indirect references of array access functions.
ATOMIZE_INDIRECT_REF_ONLY FALSE
By default, simple array accesses such as X(I+2) are atomized, although it is not necessary to generate assembly code:
ATOMIZE_ARRAY_ACCESSES_WITH_OFFSETS TRUE
The purpose of the default option is to maximise common subexpression elimination.
Once a code has been atomized, you can use this transformation to generate two address code only It can be useful for asm generation
GENERATE_TWO_ADDRESSES_CODE_SKIP_DEREFERENCING TRUE
Function partial_eval 9.4.2 produces code where numerical constant expressions or subexpressions are replaced by their value. Using the preconditions, some variables are evaluated to a integer constant, and replaced wherever possible. They are not replaced in user function calls because Fortran uses a call-by-reference mechanism and because they might be updated by the function. For the same conservative reason, they are not replaced in intrinsics calls.
Note that symbolic constants were left unevaluated because they already are constant. However it was found unfriendly by users because the principle of least surprise was not enforced: symbolic constants were sometimes replaced in the middle of an expression but not when the whole expression was a reference to a symbolic constant. Symbolic integer constants are now replaced by their values systematically.
Transformations simplify_control 9.3.1 and dead_code_elimination 9.3.2 should be performed after partial evaluation. It is sometimes important to run more than one partial evaluation in a row, because the first partial evaluation may linearize some initially non-linear expressions. Perfect Club benchmark ocean is a case in point.
Comments from Nga Nguyen: According to [?] and [?], the name of this optimization should be Constant-Expression Evaluation or Constant Folding for integer values. This transformation produces well error message at compile time indicating potential error such as division by zero.
PIPS3 default behavior in various places is to evaluate symbolic constants. While meaningful, this approach is not source-to-source compliant, so one can set property EVAL_SYMBOLIC_CONSTANT 9.4.2 to FALSE to prevent some of those evaluations.
EVAL_SYMBOLIC_CONSTANT TRUE
One can also set PARTIAL_EVAL_ALWAYS_SIMPLIFY 9.4.2 to TRUE in order to force distribution, even when it does not seem profitable
PARTIAL_EVAL_ALWAYS_SIMPLIFY FALSE
Likewise, one can turn following property to true if he wants to use hard-coded value for size of types
EVAL_SIZEOF FALSE
This function was implemented initially by Bruno Baron.
Phase Reductions detects generalized instructions and replaces them by calls to a run-time library supporting parallel reductions. It was developed by Pierre Jouvelot in CommonLISP, as a prototype, to show than NewGen data structures were language-neutral. Thus it by-passes some of pipsmake/dbm facilities.
This phase is now obsolete, although reduction detection is critical for code restructuring and optimization... A new reduction detection phase was implemented by Fabien Coelho. Have a look at § 6.4 but it does not include a code transformation. Its result could be prettyprinted in an HPF style (FC: implementation?).
replace_reduction_with_atomic 9.4.4 replace all reduction in loop that are marked as parallel with reduction by coarse_grain_parallelization_with_reduction 8.1.6.
The property ATOMIC_OPERATION_PROFILE 9.4.4 control the set of atomic operations and operand allowed. At that time only “cuda” is supported.
flag_parallel_reduced_loops_with_atomic 9.4.4 flag as parallel all loops that were detected by coarse_grain_parallelization_with_reduction 8.1.6.
The property ATOMIC_OPERATION_PROFILE 9.4.4 control the set of atomic operations and operand allowed. At that time only “cuda” is supported.
ATOMIC_OPERATION_PROFILE "cuda"
Flag loops with openmp directives, taking into account reductions.
Scalars can be forward substituted. The effect is to undo already performed optimizations such as invariant code motion and common subexpression elimination, or manual atomization. However we hope to do a better job automatically!
One can set FORWARD_SUBSTITUTE_OPTIMISTIC_CLEAN 9.4.5 to TRUE in order to clean (without check) forward - substituted assignments. Use cautiously !
FORWARD_SUBSTITUTE_OPTIMISTIC_CLEAN FALSE
This transformation is quickly developed to fulfill the need of a simple pattern matcher in pips. The user provide a module name through EXPRESSION_SUBSTITUTION_PATTERN 9.4.6 property and all expression similar to those contained in EXPRESSION_SUBSTITUTION_PATTERN 9.4.6 will be substituted to a call to this module. It is a kind of simple outlining transformations, it proves to be useful during simdization to recognize some idioms. Note that the pattern must contain only a single return instruction!
This phase was developed by Serge Guelton during his PhD.
Set RELAX_FLOAT_ASSOCIATIVITY 9.4.6 to TRUE if you want to consider all floating point operations as really associative4 :
RELAX_FLOAT_ASSOCIATIVITY FALSE
This property is used to set the one-liner module used during expression substitution. It must be the name of a module already loaded in pips and containing only one return instruction (the instruction to be matched).
EXPRESSION_SUBSTITUTION_PATTERN ""
This transformation replaces all language operators by function calls.
The function name is derived from the operator name, the operator arguments type(s) and a common prefix. Each function name is built using the pattern [PREFIX][OP NAME][SUFFIX] (eg: int + int will lead to op_addi). The replacement function must have been declared, otherwise a warning is emited and the operator is ignored.
OP NAME is defined by the following table:
| post++ | post_inc |
| ++pre | inc_pre |
| post-- | post_dec |
| --pre | dec_pre |
| + | plus |
| unary + | un_plus |
| - | minus |
| unary - | un_minus |
| * | mul |
| / | div |
| % | mod |
| = | assign |
| *= | mul_up |
| /= | div_up |
| %= | mod_up |
| += | plus_up |
| -= | minus_up |
| <= | leq |
| < | lt |
| >= | geq |
| > | gt |
| == | eq |
| != | neq |
Using the property RENAME_OPERATOR_OPS 9.4.7, it is possible to give a restrictive list of operator names on which operator renaming should be applied. Operator that are not in this list are ignored.
RENAME_OPERATOR_OPS "plus␣minus␣mul␣div␣mod␣un_plus␣un_minus␣assign␣mul_up␣div_up␣mod_up␣plus_up␣minus_up"
Assuming that all arguments of the operator have the same type. SUFFIX is deduced using the following table:
| char | c |
| short | s |
| int | i |
| long | l |
| float | f |
| double | d |
| _Bool | b |
| _Complex | C |
| _Imaginary | I |
Using the property RENAME_OPERATOR_SUFFIXES 9.4.7, it is possible to give a restrictive list of suffix on which operator renaming should be applied. Every type not listed in this list will be ignored.
RENAME_OPERATOR_SUFFIXES "f␣d␣C␣I"
The PREFIX is a common prefix defined by the property RENAME_OPERATOR_PREFIX 9.4.7 which is applied to each operators. It can be used to choose between multiple implementations of the same operator. The default value is op_.
RENAME_OPERATOR_PREFIX "op_"
In Pips, C For loop like for(i=0; i < n; i++) is represented by a Fortran-like range-based Do loop do i = 1,n-1. Thus, the code:
will be rewritten :
If you want it to be rewritten :
you should set the property RENAME_OPERATOR_REWRITE_DO_LOOP_RANGE 9.4.7 to TRUE. This is not the default behaviour, because in most case you don’t want to rewrite For loop like this.
RENAME_OPERATOR_REWRITE_DO_LOOP_RANGE FALSE
Some operators (=, +=, …) takes a modifiable lvalue. In this case, the expected function signature for a type T is T (T*, T). For instance, the code:
would be rewritten:
This transformation replaces all arrays in the module by equivalent linearized arrays. Eventually using array/pointer equivalence.
This transformation replaces all arrays in the module by equivalent linearized arrays. This only makes the arrays starting their index from one.
Use LINEARIZE_ARRAY_USE_POINTERS 9.4.8 to control whether arrays are declared as 1D arrays or pointers. Pointers are accessed using dereferencement and arrays using subscripts. This property does not apply to the fortran case.
LINEARIZE_ARRAY_USE_POINTERS FALSE
Use LINEARIZE_ARRAY_MODIFY_CALL_SITE 9.4.8 to control whether the call site is modified or not.
LINEARIZE_ARRAY_MODIFY_CALL_SITE TRUE
Use LINEARIZE_ARRAY_CAST_AT_CALL_SITE 9.4.8 to control whether a cast is inserted at call sites. Turning it on break further effects analysis, but without the cast it might break compilation or at least generate warnings for type mismatch. This property does not apply to the fortran case.
LINEARIZE_ARRAY_CAST_AT_CALL_SITE FALSE
Use LINEARIZE_ARRAY_SKIP_STATIC_LENGTH_ARRAYS 9.4.8 to skip the array to pointer conversion for static length arrays. Linearization is always done.
LINEARIZE_ARRAY_SKIP_STATIC_LENGTH_ARRAYS FALSE
Use LINEARIZE_ARRAY_SKIP_LOCAL_ARRAYS 9.4.8 to skip the array to pointer conversion for locally declared arrays. Linearization is always done.
LINEARIZE_ARRAY_SKIP_LOCAL_ARRAYS FALSE
This is an experimental section developed by Julien Zory as PhD work [?]. This phase aims at optimizing expression evaluation using algebraic properties such as associativity, commutativity, neutral elements and so forth.
This phase restructure arithmetic expressions in order (1) to decrease the number of operations (e.g. through factorization), (2) to increase the ILP by keeping the corresponding DAG wide enough, (3) to facilitate the detection of composite instructions such as multiply-add, (4) to provide additional opportunities for (4a) invariant code motion (ICM) and (4b) common subexpression elimination (CSE).
Large arithmetic expressions are first built up via forward substitution when the programmer has already applied ICM and CSE by hand.
The optimal restructuring of expressions depends on the target defined by a combination of the computer architecture and the compiler. The target is specified by a string property called EOLE_OPTIMIZATION_STRATEGY 9.4.9 which can take values such as "P2SC" for IBM Power-2 architecture and XLF 4.3. To activate all sub-transformations such as ICM and CSE set it to "FULL". See properties for more information about values for this property and about other properties controlling the behavior of this phase.
The current implementation is still shaky and does not handle well expressions of mixed types such as X+1 where 1 is implictly promoted from integer to real.
Warning: this phase relies on an external (and unavailable) binary. To make it work, you can set EOLE_OPTIMIZATION_STRATEGY 9.4.9 to "CSE" or "ICM", or even ICMCSE to have both. This will only activate common subexpressions elimination or invariant code motion. Since it is a quite common use case, they have been defined as independent phase too. See 9.4.10.
EOLE: Evaluation Optimization of Loops and Expressions. Julien Zory stuff integrated within pips [?]. It relies on an external tool named eole. The version and options set can be controlled from the following properties. The status is experimental. See the optimize_expressions 9.4.9 pass for more details about the advanced transformations performed.
EOLE "newgen_eole"
EOLE_FLAGS "-nfd"
EOLE_OPTIONS ""
EOLE_OPTIMIZATION_STRATEGY "P2SC"
Here are described two interesting cases of the one in § 9.4.9.
Run common sub-expression elimination to factorize out some redundant expressions in the code.
One can use COMMON_SUBEXPRESSION_ELIMINATION_SKIP_ADDED_CONSTANT 9.4.10 to skip expression of the form a+2 and COMMON_SUBEXPRESSION_ELIMINATION_SKIP_LHS 9.4.10 to prevent elimination of left hand side of assignment.
The heuristic used for common subexpression elimination is described in Chapter 15 of Julien Zory’s PhD dissertation [?].
Note: the icm deals with expressions while the invariant_code_motion deals with loop invariant code.
The following property is used in sac to limit the subexpressions: When set to true, only subexpressions without ”+constant” terms are eligible.
COMMON_SUBEXPRESSION_ELIMINATION_SKIP_ADDED_CONSTANT FALSE
COMMON_SUBEXPRESSION_ELIMINATION_SKIP_LHS TRUE
Performs invariant code motion over sub expressions
Generate code from a FREIA application possibly targeting hardware accelerator, such as SPoC, Terapix, or GPGPU. I’m unsure about the right granularity (now it is at the function level) and the resource which is produced (should it be an accelerated file?). The current choice does not allow to easily mix different accelerators.
Generate code for a software FREIA implementation, by applying various optimizations at the library API level, but without generating accelerated functions.
The following properties are generic to all FREIA accelerator targets.
Whether to label arcs in dag dot output with the image name, and to label nodes with the statement number, and whether to filter out unused scalar nodes.
FREIA_DAG_LABEL_ARCS FALSE
FREIA_DAG_LABEL_NODES TRUE
FREIA_DAG_FILTER_NODES TRUE
Whether to compile lone operations, i.e. operations which do not belong to a sequence.
FREIA_COMPILE_LONE_OPERATIONS TRUE
Whether to normalize some operations:
FREIA_NORMALIZE_OPERATIONS TRUE
Whether to simplify the DAG using algebraic properties.
FREIA_SIMPLIFY_OPERATIONS TRUE
Whether to remove dead image operations in the DAG. Should always be beneficial.
FREIA_REMOVE_DEAD_OPERATIONS TRUE
Whether to remove duplicate operations in the DAG, including algebraic optimizations with commutators. Should be always beneficial to terapix, but it may depend for spoc.
FREIA_REMOVE_DUPLICATE_OPERATIONS TRUE
Whether to remove useless image copies from the expression DAG.
FREIA_REMOVE_USELESS_COPIES TRUE
Whether to move image copies within an expression DAG outside as external copies, if possible.
FREIA_MOVE_DIRECT_COPIES TRUE
Whether to merge identical arguments, especially kernels, when calling an accelerated function:
FREIA_MERGE_ARGUMENTS TRUE
Whether to attempt to reuse initial images if possible, instead of keeping possibly newly introduced temporary images.
FREIA_REUSE_INITIAL_IMAGES TRUE
Try to allow shuffling image pointers, but this is not allowed by default because it may lead to wrong code as the compiler currently ignores the information and mixes up images.
FREIA_ALLOW_IMAGE_SHUFFLE FALSE
Whether to assume that casts are simple image copies. Default is to keep a cast as cast, which is not accelerated.
FREIA_CAST_IS_COPY FALSE
Whether to cleanup freia returned status, as the code is assumed correct when compiled.
FREIA_CLEANUP_STATUS TRUE
Assume this pixel size in bits:
FREIA_PIXEL_SIZE 16
If set to a non-zero value, assume this image size when generating code. If zero, try generic code. In particular, the height is useful to compute a better imagelet size when generating code for the Terapix hardware accelerator.
FREIA_IMAGE_HEIGHT 0
FREIA_IMAGE_WIDTH 0
Ad-hoc transformation to remove particular scalar write-write dependencies in sequences. They are introduced by the do-while to while conversion on FREIA convergence loops. There is an underlying generic transformation on sequences that could be implemented with more thoughts on the subject.
FREIA Compiler for SPoC target.
Consider applying freia_unroll_while beforehand to unroll convergence use with the right number of iterations to make the best use of the available hardware. Note that the same transformation would also make sense somehow on sequences when the do-while to while transformation as been applied, but the unrolling factor is much harder to decide in the sequence case as it would depend on previous operations.
Default depth of the target SPoC accelerator:
HWAC_SPOC_DEPTH 8
FREIA compiler for Terapix target.
Number of processing elements (PE) for the Terapix accelerator:
HWAC_TERAPIX_NPE 128
Default size of memory, in pixel, for the Terapix accelerator (RAMPE is RAM of PE):
HWAC_TERAPIX_RAMPE 1024
Terapix DMA bandwidth. How many terapix cycles to transfer an imagelet row (size of which is necessarily the number of pe):
HWAC_TERAPIX_DMABW 32
Terapix 2D global RAM (GRAM) width and height:
HWAC_TERAPIX_GRAM_WIDTH 64
HWAC_TERAPIX_GRAM_HEIGHT 32
Whether and how to further cut the dag for terapix. Expected values, by order of compilation costs, are: none, compute, enumerate.
HWAC_TERAPIX_DAG_CUT "compute"
Whether input and output memory transfers overlap one with the other, that we have a full duplex DMA.
HWAC_TERAPIX_OVERLAP_IO FALSE
Note that it is already assumed that computations overlap with communications. This adds the information that host-accelerator loads and stores run in parallel. This has two impacts: the communication apparent time is reduced thanks to the overlapping, which is good, but the imagelet memory cannot be reused for inputs because it is still live while being stored, which is bad for memory pressure.
Use this maximum size (height) of an imagelet if set to non-zero. It may be useful to set this value to the image height (if known) so that compiler generates code for smaller imagelets, so that the runtime is not surprised. This is rather use of debug to impose an imagelet size.
HWAC_TERAPIX_IMAGELET_MAX_SIZE 0
FREIA compiler for OpenCL target, which may run on both multi-core and GPU.
Whether we should attempt to generate merged OpenCL image operations. If not, the result will be simular to simple AIPO compilation, no actual helper functions will be generated.
HWAC_OPENCL_MERGE_OPERATIONS TRUE
Whether merged OpenCL image operations should include reductions as well. Added to help debugging the code.
HWAC_OPENCL_MERGE_REDUCTIONS TRUE
Whether to generate OpenCL code for one operation on its own. It is not interesting to do so because it is just equivalent to the already existing AIPO implementation, but it can be useful for debug.
HWAC_OPENCL_COMPILE_ONE_OPERATION FALSE
Inlining is a well known technique. Basically, it replaces a function call by the function body. The current implementation does not work if the function has static declarations, access global variables …. Actually it (seems to) work(s) for pure, non-recursive functions …and not to work for any other kind of call.
Property INLINING_CALLERS 9.6.1 can be set to define the list of functions where the call sites have to be inlined. By default, all call sites of the inlined function are inlined.
Only for C because of pipsmake output declaration !
Use following property to control how generated variables are initialized
INLINING_USE_INITIALIZATION_LIST TRUE
Use following property to control whether inlining should ignore stubs:
INLINING_IGNORE_STUBS TRUE
Set the following property to TRUE to add comments on inlined statements to keep track of their origin.
INLINING_COMMENT_ORIGIN FALSE
Same as inlining but always simulate the by-copy argument passing
Only for C because of pipsmake output declaration !
Regenerate the ri from the ri ...
Only for C because of pipsmake output declaration !
The default behavior of inlining is to inline the given module in all call sites. Use INLINING_CALLERS 9.6.1 property to filter the call sites: only given module names will be considered.
INLINING_CALLERS ""
Unfolding is a complementary transformation of inlining 9.6.1. While inlining inlines all call sites to a given module in other modules, unfolding inlines recursively all call sites in a given module, thus unfolding the content of the module. An unfolded source code does not contain any call anymore. If you run it recursievly, you should set INLINING_USE_INITIALIZATION_LIST 9.6.1 to false.
Only for C because of output declaration in pipsmake rule!
Same as unfolding, but cumulated effects are not used, and the resulting code always simulates the by-copy argument passing.
Only for C because of output declaration in pipsmake rule!
Use UNFOLDING_CALLEES 9.6.2, to specify which modules you want to inline in the unfolded module. The unfolding will be performed as long as one of the module in UNFOLDING_CALLEES 9.6.2 is called. More than one module can be specified, they are separated by blank spaces.
UNFOLDING_CALLEES ""
The default behavior of the unfolding 9.6.2 pass is to recursively inline all callees from the current module or from the argument modules of the pass, as long as a callee remains. You can use UNFOLDING_FILTER 9.6.2 to inline all call sites to a module not present in the space separated module list defined by the string property:
UNFOLDING_FILTER ""
By default this list is empty and hence all call sites are inlined.
This documentation is a work in progress, as well as the documented topicSerge? Still true? What do you mean?.
Outlining is the opposite transformation of inlining 9.6.1. It replaces some statements in an existing module by a call site to a new function whose execution is equivalent to the execution of the replaced statements. The body of the new function is similar to the piece of code replaced in the existing module. The user is prompted for various pieces of information in order to perform the outlining:
The statements are a subset of a sequence. They are counted in the sequenceSerge: I invent...(FI).
OUTLINE_WRITTEN_SCALAR_BY_REFERENCE 9.6.3 controls whether we pass written scalar by reference or not. This property might lead to incorrect code !
The property OUTLINE_SMART_REFERENCE_COMPUTATION 9.6.3 is used to limit the number of entities passed by reference. With it, a[0][0] is passed as an a[n][m] entity, without it it is passed as an int or int* depending on the cumulated read/write memory effects of the outlined statementsSerge: but cumulated effects are always required?.
If you need to pass the upper bound expression of a particular loop as a parameter, which is used in Ter@pix code generation (see Section ???), set OUTLINE_LOOP_BOUND_AS_PARAMETER 9.6.3 to the loop labelSerge: a few words of motivation?.
The property OUTLINE_MODULE_NAME 9.6.3 is used to specify the new module name. The user is prompted if it is set to its default value, the empty string.
But first the pass scansSerge: unconditionally? the code for any statement flagged with the pragma defined by the string property OUTLINE_PRAGMA 9.6.3Serge: What happens when it is set to the empty strin?.
If set, the string property OUTLINE_LABEL 9.6.3 is used to choose the statementSerge: you outline a set of statements; is it reduced to a singleton in that case? to outline.
The boolean property OUTLINE_ALLOW_GLOBALS 9.6.3 controls whether global variables whose initial values are not used are passed as parameters or notSerge: why is it called “globals”? It seems that this makes sense for any local variable.... It is suggested to addressed this issue with a previous privatization pass.
Finally, the boolean property OUTLINE_INDEPENDENT_COMPILATION_UNIT 9.6.3 can be set to true to outline the new module into a newly created compilation unit. It is named after the OUTLINE_MODULE_NAME 9.6.3. All necessary types, global variables and functionsSerge? are declared into this new compilation unitSerge: Consistency betweeen properties? Name conflicts?. All the functions brought in the new compilation unit through the sub-callgraph are declared static and prefixed with OUTLINE_CALLEES_PREFIX 9.6.3.
OUTLINE_MODULE_NAME ""
OUTLINE_PRAGMA "pips␣outline"
OUTLINE_LABEL ""
OUTLINE_ALLOW_GLOBALS FALSE
OUTLINE_SMART_REFERENCE_COMPUTATION FALSE
OUTLINE_LOOP_BOUND_AS_PARAMETER ""
OUTLINE_INDEPENDENT_COMPILATION_UNIT FALSE
OUTLINE_WRITTEN_SCALAR_BY_REFERENCE TRUE
OUTLINE_CALLEES_PREFIX ""
Procedures can be cloned to obtain several specialized versions. The call sites must be updated to refer to the desired version.
User assisted cloning.See examples in clone validation suite. RK: terse; to be improved by FC
Cloning of a subroutine according to an integer scalar argument. The argument is specified through integer property TRANSFORMATION_CLONE_ON_ARGUMENT 9.6.4. If set to 0, a user request is performed.
Not use assisted version of cloning it just perform the cloning without any substitution Use the CLONE_NAME 9.6.4 property if you want a particular clone name. It’s up to another phase to perform the substitution.
There are two cloning properties. Cloning on an argument. If 0, a user request is performed.
TRANSFORMATION_CLONE_ON_ARGUMENT 0
Clone name can be given using the CLONE_NAME properties Otherwise, a new one is generated
CLONE_NAME ""
Clean the declarations of unused variables and commons and so. It is also a code transformation, since not only the module entity are updated by the process, but also the declaration statements, some useless writes...
Clean the declarations of unused variables and commons and so.
In C, dynamic variables which are allocated and freed but otherwise never used can be removed. This phase removes the calls to the dynamic allocation functions (malloc and free or user defined equivalents), and remove their declarations.
Clean unused local dynamic variables by removing malloc/free calls.
It may be a regular expression instead of a function name?
DYNAMIC_ALLOCATION "malloc"
DYNAMIC_DEALLOCATION "free"
Detecting and forcing variables in the register storage class can help subsequent analyses, as they cannot be referenced by pointers.
Whether to allow arrays to be qualified as registers:
FORCE_REGISTER_ARRAY TRUE
Whether to allow pointers to be qualified as registers:
FORCE_REGISTER_POINTER TRUE
Whether to allow formal parameters to be qualified as registers:
FORCE_REGISTER_FORMAL TRUE
One problem of Fortran code is the unnormalized array bound declarations. In many program, the programmer put an asterisk (assumed-size array declarator), even 1 for every upper bound of last dimension of array declaration. This feature affects code quality and prevents others analyses such as array bound checking or alias analysis. We developed in PIPS two new methods to find out automatically the proper upper bound for the unnormalized and assumed-size array declarations, a process we call array resizing. Both approaches have advantages and drawbacks and maybe a combination of these ones is needed.
To have 100% resized arrays, we implement also the code instrumentation task, in the top-down approach.
Different options to compute new declarations for different kinds of arrays are described in properties-rc.tex. You can combine the two approaches to have a better results by using these options.
How to use these approaches: after generating new declarations in the logfile, you have to use the script $PIPS_ROOT/Src/Script/misc/array_resizing_instrumentation.pl to replace the unnormalized declarations and add new assignments in the source code.
The method uses the relationship between actual and formal arguments from parameter-passing rules. New array declarations in the called procedure are computed with respect to the declarations in the calling procedures. It is faster than the first one because convex array regions are not needed.
This phase is implemented by Thi Viet Nga Nguyen (see [?]).
The approach is based on an convex array region analysis that gives information about the set of array elements accessed during the execution of code. The regions READ and WRITE of each array in each module are merged and a new value for the upper bound of the last dimension is calculated and then it will replace the 1 or *.
This function is firstly implemented by Trinh Quoc Anh, and ameliorated by Corinne Ancourt and Thi Viet Nga Nguyen (see [?]).
We provide here a tool to calculate the number of pointer-type A(,1) and assumed-size A(,*) array declarators as well as other information.
This phase is firstly designed to infer automatically new array declarations for assumed-size (A(*)) and one (A(1) or also called ugly assumed-size) array declarators. But it also can be used for all kinds of array : local or formal array arguments, unnormalized or all kinds of declarations. There are two different approaches that can be combined to have better results.
There are three different options:
So the combination of the three above options gives us a number from 0 to 7 (binary representation : 000, 001,..., 111). You must pay attention to the order of options. For example, if you want to use information from MAIN program to compute new declarations for assumed-size and one array declarations, both of them, the option is 4 (100). The default option is 0 (000).
ARRAY_RESIZING_TOP_DOWN_OPTION 0
There are also three different options:
So the combination of the three above options gives us a number from 0 to 7 (binary representation : 000, 001,..., 111). You must pay attention to the order of options. There are some options that exclude others, such as the option to compute new declarations for instrumented array (I_PIPS_MODULE_ARRAY). The default option is 0 (000).
ARRAY_RESIZING_BOTTOM_UP_OPTION 0
Scalarization is the process of replacing array references with references to scalar copies of the array elements wherever appropriate. Expected benefits include lower memory footprint and access time because registers can be used instead of temporary stack variables or memory accesses, and hence, shorter execution times, as long as the register pressure is not so high that spill code is generated.
Scalarizing a given array reference is subject to two successive criteria, a Legality criterion and a Profitability criterion:
The legality test is based on convex array regions, and not on the dependence graph as is the Carr’s algorithm. Currently, loop carried dependence arcs prevent scalarization by PIPS, although some can be processed by Carr’s algorithm. See non-regression tests Transformations/scalarization30 to 36.
This transformation is useful 1) to improve the readability of the source code with constructs similar to let x be ..., 2) to improve the modelizations of the execution, time or energy, by using source instructions closer to the machine instruction, 3) to perform an optimization at the source level because the code generator does not include a powerful (partially) redundant load elimination, and 4) to be a useful pass in a fully source-to-source compiler. This transformation is useful to reduce the expansion caused by the different atomizer passes.
The new scalar variables use the default prefix ___scalar__ and are thus easily identified, but a new prefix can be user defined with Property SCALARIZATION_PREFIX 9.7.3. IfSCALARIZATION_PREFIX 9.7.3 is the empty string, the names of the scalarized variables are used.
If needed according to the IN and OUT convex array regions, the new variables are initialized, e.g. __scalar0__ = A[i], and/or copied back into the initial array, e.g. A[i] = __scalar0__.
Scalarization is currently applicable both to Fortran and C code, but the code generation slightly differs because local declarations are possible with C. However, this may destroy perfect loop nests required by other passes. Property SCALARIZATION_PRESERVE_PERFECT_LOOP_NEST 9.7.3 can be used to preserve C perfect loop nests, this property is currently not completely implemented. SCALARIZATION_KEEP_PERFECT_PARALLEL_LOOP_NESTS 9.7.3 can be used to preserve C perfect parallel loop nests.
Pass scalarization 9.7.3 uses the read and written convex array regions to decide if the scalarization is possible, the IN and OUT regions to decide if it is useful to initialize the scalar copy or to restore the value of the array element.
Since OUT regions are computed interprocedurally, strange code may result when a small or not so small function without any real output is scalarized. The problem can be fixed by adding a PRINT or a printf to force an OUT region, or by setting Property SCALARIZATION_FORCE_OUT 9.7.3 to true.
Also, it may be useful to use Property SEMANTICS_TRUST_ARRAY_DECLARATIONS 6.9.4.2 and/or SEMANTICS_TRUST_ARRAY_REFERENCES 6.9.4.2 to make sure that as many loops as possible are entered. If not, loop bounds may act like guards, which prevents PIPS from hoisting a reference out of a loop, although this is puzzling for programmers because they expect all loops to be entered at least once...
As explained above, Property SCALARIZATION_PREFIX 9.7.3 is used to select the names of the new scalar variables. If it is set to the empty string, "", the scalarized variable name is used as prefix, which improves readability.
SCALARIZATION_PREFIX ""
When property SCALARIZATION_USE_REGISTERS 9.7.3 is set to TRUE (default value), new variables are declared with the register qualifier. However, this is not always compatible with other phases. Set it to FALSE to toggle off this behavior.
SCALARIZATION_USE_REGISTERS TRUE
Also, as explained above, Property SCALARIZATION_PRESERVE_PERFECT_LOOP_NEST 9.7.3 is used to control the way new scalar variables are declared and initialized in C. The current default value is FALSE because the locality of declarations is better for automatic loop parallelization, but this could be changed as a privatization pass should do as well while preserving perfect loop nests. Note that initializations may have to be placed in such a way that a perfect loop nest is nevertheless destroyed, even with this property set to true.
SCALARIZATION_PRESERVE_PERFECT_LOOP_NEST FALSE
SCALARIZATION_KEEP_PERFECT_PARALLEL_LOOP_NESTS FALSE
Property SCALARIZATION_FORCE_OUT 9.7.3 forces the generation of a copy-out statement when it is useless according to the OUT region, for instance when dealing with a library function.
SCALARIZATION_FORCE_OUT FALSE
Numerical property SCALARIZATION_THRESHOLD 9.7.3 is used to decide profitability: the estimated complexity of the scalarized code must be less than the initial complexity. Its minimal value is 2. It can be set to around 5 to forbid scalarization in sequences with no loop benefit.
SCALARIZATION_THRESHOLD 2
Property SCALARIZATION_ACROSS_CONTROL_TEST 9.7.3 is used to control the place where new memory accesses are inserted.
A memory access may be moved out of a loop, which is the best case for performance, but it then is no longer control dependent on the loop bounds. A new access is thus added if the compiler cannot prove that the loop is always entered. However, the burden of the proof may be too much, if only because nothing can be proven when the loop is sometimes entered and sometimes not. Also, the memory access may be always perfectly legal.
The default value for SCALARIZATION_ACROSS_CONTROL_TEST 9.7.3 is "exactness", allying safety to performance. It forbids scalarization when both read and written regions for the current piece of code are approximate regions, and, in addition, it applies a cheap test. This is the default option.
When it is set to "strict", property SCALARIZATION_ACROSS_CONTROL_TEST 9.7.3 goes for safety before performance, allowing candidates from approximated regions, but performing a strict test to check the validity of the transformation.
When the property is set to "cheap", memory accesses are moved outside of control structures with only the cheapest test. Use this last value with caution, since it is unsafe in the general case.
These properties are pretty hard to understand, as well as their impact. See sequence06, 07, 08 and 44 in the validation suite of the scalarization pass.
SCALARIZATION_ACROSS_CONTROL_TEST "exactness"
This pass was designed by François Irigoin and implemented by Laurent Daverio and François Irigoin.
Similar to scalarization 9.7.3, but with a different criterion: if an array is only accessed with numerical constant subscript expressions, it is replaced by a set of scalars and all its references are replaced by references to the corresponding scalars.
This pass may be useful to make C source code more palatable to C to VHDL converters that handle scalars better than arrays. It is also useful to propagate constants contained in an array since the semantics passes only analyze scalar variables.
This pass was designed and implemented by Serge Guelton.
Pass scalarization 9.7.3 may be costly because it relies on array regions analyses and uses them to check its legality criteria. The next phase alleviates these drawbacks.
This pass solely relies on proper and cumulated effets, and as such may fail to scalarize some accesses. However, it is expected to give good results in usual cases, especially after loop_fusion 9.1.6.
It basically uses the same algorithm as scalar privatization, but performs on the dependence graph rather than the chains graph for more precision about array dependences. Several legality criteria are then tested to ensure the safety of the transformation. In particular, it is checked that candidate references are not accessed through hidden references (for instance in calls), and that only one kind of reference is scalarized in the loop.
Pass quick_scalarization 9.7.3 was implemented by Beatrice Creusillet.
Induction substitution is the process of replacing scalar variables by a linear expression of the loop indices.
This pass was designed and implemented by Mehdi Amini.
Reduce the complexity of expression computation by generating induction variables when possible. E.g.
Would become
The goal of this program transformation is to enlarge basic blocks as much as possible to increase the opportunities for optimization. The input code is assumed serial: parallel loops are declared sequential.
This transformation has been developed in PIPS for heterogeneous computing and is combined with inlining to increase the size of the code executed by an external accelerator while reducing the externalization overhead5 . Other transformations, such as partial evaluation and dead code elimination (including use-def elimination) can be applied to streamline the resulting code further.
The transformation flatten_code 9.7.6 firstly moves declarations up in the abstract syntax tree, secondly remove useless braces and thirdly fully unroll loops when there iteration counts are known and the FLATTEN_CODE_UNROLL property is true. Unrolling can also be controlled using Property FULL_LOOP_UNROLL_EXCEPTIONS 9.1.8.2. The declarations are moved up: some parallel loops may become sequential. To avoid inconsistency, all loops are declared sequential by flatten_code 9.7.6.
Inlining(s), which must be performed explicitly by the user with tpips or another PIPS interface, can be used first to create lots of opportunities. The basic block size increase is first due to brace removals made possible when declarations have been moved up, and then to loop unrollings. Finally, partial evaluation, dead code elimination and use-def based elimination can also straighten-out the code and enlarge basic blocks by removing useless tests or assignments.
The code externalization and adaptation for a given hardware accelerator is performed by another phase, see for instance Section 9.5.
Initially developed in August 2009 by Laurent Daverio, with help from Fabien Coelho and Francois Irigoin.
If the following property is set, loop unrolling is applied too for loops with static bounds.
FLATTEN_CODE_UNROLL TRUE
Split C operators such as a += b, a *= b, a >>= b, etc. into their expanded form such as a = a + b.
Note that if the left hand side expression lhs implies side effects, the transformed code is not equivalent since lhs be evaluated twice in the transformed code. The left hand side is not checked for side effects. The legality of the transformation is not guaranteed.
Two improvements could be used. Check the side effects with a function such as expression_to_proper_constant_path_effects() and/or use a pointer to evaluate lhs a once to obtain p = &a; *p = *p +b;.
The purpose of this transformation is to separate the initialization part from the declaration part in C code in order to make static code analyses simpler.
This transformation recurses through all variable declarations, and creates a new statement each time an initial value is specified in the declaration, if the initial value can be assigned. The declarations are modified by eliminating the initial value, and a new assignment statement with the initial value is added to the source code.
This transformation can be used, for instance, to improve reduction detection (see TRAC Ticket 181).
Note that C array and structure initializations, which use braces, cannot be converted into assignments. In such cases, the initial declaration is left untouchedFI-¿SG: improved?.
This transformation uses the C89_CODE_GENERATION property to generate either C89 or C99 code.
The purpose of this transformation is to change the return type of a function. The new type will be a typedef whose name is controlled by SET_RETURN_TYPE_AS_TYPEDEF_NEW_TYPE 9.7.9. The corresponding typedef must exist in the symbol table.
This transformation loops over the symbols in the symbol table, and for each of them which is a typedef, compare the local name to the property SET_RETURN_TYPE_AS_TYPEDEF_NEW_TYPE 9.7.9. This approach is unsafe because there can be different typedef with the same name in different compilation units, resulting in differents entries in the symbol table for a same local name. The return type can also be incoherent with the return statement, thus it is not safe to run it on a non-void function.
However this pass has been created for special need in par4all, and considering restrictions described above, it does the job.
SET_RETURN_TYPE_AS_TYPEDEF_NEW_TYPE "P4A_accel_kernel_wrapper"
The purpose of this transformation is to cast parameters at call sites according to the prototype, a.k.a. the signature, of the called functions.
Variable privatization consists in discovering variables whose values are local to a particular scope, usually a loop iteration.
Three different privatization functions are available. The quick privatization is restricted to loop indices and is included in the dependence graph computation (see Section 6.6). The scalar privatization should be applied before any serious parallelization attempt. The array privatization is much more expensive and is still mainly experimental.
You should keep in mind that, although they modify the internal repesentation of the code, scalar and array privatizations are only latent program transformations, and no actual local variable declaration is generated. This is the responsibility of code generation phases, which may use this information differently depending on their target.
Privatizer detects variables that are local to a loop nest and marks these variables as private. A variable is private to a loop if the values assigned to this variable inside the loop cannot reach a statement outside the loop body.
Note that illegal code, for instance code with uninitialized variables, can lead to surprising privatizations, which are still correct since the initial semantics is unspecified.
This pass is similar to privatize_module 9.7.11.1, but it also privatizes global scalar variables, using information from live_paths 6.3 analyses to avoid privatizing global variables which are the values of which are used afterwards, and callees summary_effects 6.2.4 to ensure that global variables are not used in callees. This last property is necessary to keep the transformation as local as possible. If it were not true, it would require to clone the sub call-trees.
Use informations from privatize_module 9.7.11.1 to move C variable declarations as close as possible to their uses. For instance
becomes
LOCALIZE_DECLARATION_SKIP_LOOP_INDICES FALSE
Array privatization aims at privatizing whole arrays (array_privatizer 9.7.11.3) or sets of array elements (array_section_privatizer 9.7.11.3) instead of scalar variables only. The algorithm, developed by Béatrice Creusillet [?], is very different from the algorithm used for solely privatizing scalar variables and relies on IN and OUT regions analyses. Of course, it also privatizes scalar variables, although the algorithm is much more expensive and should be used only when necessary.
Array sections privatization is still experimental and should be used with great care. In particular, it is not compatible with the next steps of the parallelization process, i.e. dependence tests and code generation, because it does not modify the code, but solely produces a new region resource.
Another transformation, which can also be called a privatization, consists in declaring as local to a procedure or function the variables which are used only locally. This happens quite frequently in old Fortran codes where variables are declared as SAVEd to avoid allocations at each invocation of the routine. However, this prevents parallelization of the loop surrounding the calls. The function which performs this transformation is called declarations_privatizer 9.7.11.3.
Several privatizability criterions can be applied for array section privatization, and its not yet clear which one should be used. The default case is to remove potential false dependences between iterations. The first option, when set to false, removes this constraint. It is useful for single assignment programs, to discover what section is really local to each iteration.
When the second option is set to false, copy-out is not allowed, which means only array regions that are not further reused in the program continuation can be privatized.
ARRAY_PRIV_FALSE_DEP_ONLY TRUE
ARRAY_SECTION_PRIV_COPY_OUT TRUE
Variable expansion consists in adding new dimensions to a variable so as to parallelize surrounding loops. There is no known advantage for expansion against privatization, but expansion is used when parallel loops must be distributed, for instance to generate SIMD code.
It is assumed that the variables to be expanded are the private variables. So this phase only is useful if a privatization has been performed earlier.
Loop private scalar variables are expanded
Uses LOOP_LABEL 9.1.1 to select a particular loop, then finds all reduction in this loop and performs variable expension on all reduction variables.
A variant of atomization that splits expressions but keep as much reduction as possible. E.g: r+=a+b becomes r+=a ; r+=b;
Not implemented yet.
Function freeze_variables 9.7.13 produces code where variables interactively specified by the user are transformed into constants. This is useful when the functionality of a code must be reduced. For instance, a code designed for N dimensions could be reduced to a 3-D code by setting N to 3. This is not obvious when N changes within the code. CA? More information? The variable names are requested from the PIPS user? This is useful to specialize a code according to specific input data6 .
The following warning paragraphs should not be located here, but the whole introduction has to be updated to take into account the merger with properties-rc.tex, the new content (the transformation section has been exploded) and the new passes such as gpips. No time right now. FI.
All PIPS transformations assume that the initial code is legal according to the language standard. In other words, its semantics is well defined. Otherwise, it is impossible to maintain a constant semantics through program transformations. So uninitialized variables, for instance, can lead to codes that seem wrong, because they are likely to give different outputs than the initial code. But this does not matter as the initial code output is undefined and could well be the new output,
Also, remember that dead code does not impact the semantics in an observable way. Hence dead code can be transformed in apparently weird ways. For instance, all loops that are part of a dead code section can be found parallel, although they are obviously sequential, because all the references will carry an unfeasible predicate. In fact, reference A(I), placed in a dead code section, does not reference the memory and does not have to be part of the dependence graph.
Dead code can crop out in many modules when a whole application linked with a library is analyzed. All unused library modules are dead for PIPS.
On the other hand, missing source modules synthesized by PIPS may also lead to weird results because they are integrated in the application with empty definitions. Their call sites have no impact on the application semantics.
Typecheck code according to Fortran standard + double-complex. Typechecking is performed interprocedurally for user-defined functions. Insert type conversions where implicitly needed. Use typed intrinsics instead of generic ones. Precompute constant conversions if appropriate (e.g. 16 to 16.0E0). Add comments about type errors detected in the code. Report back how much was done.
Here are type checker options. Whether to deal with double complex or to refuse them. Whether to add a summary of errors, conversions and simplifications as a comment to the routine. Whether to always show complex constructors.
TYPE_CHECKER_DOUBLE_COMPLEX_EXTENSION FALSE
TYPE_CHECKER_LONG_DOUBLE_COMPLEX_EXTENSION FALSE
TYPE_CHECKER_ADD_SUMMARY FALSE
TYPE_CHECKER_EXPLICIT_COMPLEX_CONSTANTS FALSE
The window interfaces let the user edit the source files, because it is very useful to demonstrate PIPS. As with stf 9.3.4, editing is not integrated like other program transformations, and previously applied transformations are lost. Consistency is however always preserved.
A general edit facility fully integrated in pipsmake is planned for the (not so) near future. Not so near because user demand for this feature is low.
Since tpips can invoque any Shell command, it is also possible to touch and edit source files.
This is plug to implement quickly a program transformation requested by a user. Currently, it is a full loop distribution suggested by Alain Darte to compare different implementations, namely Nestor and PIPS.
The following transformation reads the sequential code and generates OpenMP pragma as an extension to statements. The pragmas produced are based on the information previously computed by differents phases and already stores in the pips internal representation of the sequential code. It might be interesting to use the phase internalize_parallel_code (see § 8.1.8) before to apply ompify_code in order to maximize the number of parallel information available.
As defined in the ri, the pragma can be of different types. The following property can be set to str or expr. Obviously, if the property is set to str then pragmas would be generated as strings otherwise pragmas would be generated as expressions.
PRAGMA_TYPE "expr"
The PIPS phase OMP_LOOP_PARALLEL_THRESHOLD_SET allows to add the OpenMP if clause to all the OpenMP pragmas. Afterwards, the number of iterations in the loop is evaluated dynamically and compared to the defined threshold. The loop is parallelized only if the threshold is reached.
The OMP_LOOP_PARALLEL_THRESHOLD_VALUE property , is used as a parameter by the PIPS phase OMP_LOOP_PARALLEL_THRESHOLD_SET. The number of iteration of the parallel loop will be compared to that value in an omp if clause. The OpenMP run time will decide dynamicaly to parallelize the loop if the number of iteration is above this threshold.
OMP_LOOP_PARALLEL_THRESHOLD_VALUE 0
The OMP_IF_CLAUSE_RECURSIVE property , is used as a parameter by the PIPS phase OMP_LOOP_PARALLEL_THRESHOLD_SET. If set to TRUE the number of iterations of the inner loops will be used to test if the threshold is reached. Otherwise only the nunber of iteration of the processed loop will be used.
OMP_IF_CLAUSE_RECURSIVE TRUE
Compiler tends to produce many parallel loops which is generally not optimal for performance. The following transformation merges nested omp pragma in a unique omp pragma.
PIPS merges the omp pragma on the inner or outer loop depending on the property OMP_MERGE_POLICY. This string property can be set to either outer or inner.
OMP_MERGE_POLICY "outer"
The OMP_MERGE_PRAGMA phase with the inner mode can be used after the phase limit_nested_parallelism (see § 8.1.11). Such a combinaison allows to fine choose the loop depth you really want to parallelize with OpenMP.
The merging of the if clause of the omp pragma follows its own rule. This clause can be ignore without changing the output of the program, it only changes the program perfomances. Then three policies are offered to manage the if clause merging. The if clause can simply be ignored. Or the if clauses can be merged alltogether using the boolean opertaion or or and. When ignored, the if clause can be later regenerated using the appropriated PIPS phase : OMP_LOOP_PARALLEL_THRESHOLD_SET. To summarize, remenber that the property can be set to ignore or or and
OMP_IF_MERGE_POLICY "ignore"
PIPS results for any analysis and/or transformations can be displayed in several different formats. User views are the closest one to the initial user source code. Sequential views are obtained by prettyprinting the PIPS internal representation of modules. Code can also be displayed graphically or using Emacs facilities (through a property). Of course, parallelized versions are available. At the program level, call graph and interprocedural control flow graphs, with different degrees of ellipse, provide interesting summaries.
Dependence graphs can be shown, but they are not user-friendly. No filtering interface is available. They mainly are useful for debugging and for teaching purposes.
These are files containing a pretty-printed version of the parsed code, before the controlizer is applied. It is the code display closest to the user source code, because arcs in control flow graphs do not have to be rewritten as GOTO statements. However, it is inconsistent with the internal representation of the code as soon a a code transformation has been applied.
Bug: the inconsistence between the user view and the internal code representation presently is not detected. Solution: do not use user views.
The Fortran statements may be decorated with preconditions or transformers or complexities or any kind of effects, including convex array regions,... depending on the prettyprinter selected used to produce this file.
Transformers and preconditions require cumulated effects to build the module value basis.
Display the code without any decoration.
Display the code decorated with the transformers.
Display the code decorated with the preconditions.
Display the code decorated with the total preconditions.
Display the code decorated with the continuation conditions.
Display the code decorated with the regions.
Display the code decorated with the regions.
Display the code decorated with the IN regions.
Display the code decorated with the OUT regions.
Display the code decorated with the complexities.
Display the code decorated with the proper effects.
Display the code decorated with the cumulated effects.
Display the code decorated with its IN effects.
Display the code decorated with its OUT effects.
These are files containing a pretty-printed version of the internal representation, code.
The statements may be decorated with the result of any analysis, e.g.complexities, preconditions, transformers, convex array regions,… depending on the pretty printer used to produce this file.
To view C programs, it is a good idea to select a C pretty printer, for example in tpips with:
Transformers and preconditions (and regions?) require cumulated effects to build the module value basis.
This is intended to be used with PIPS IR Navigator (tm).
Produce a html version of the internal representation of a PIPS Module. The property HTML_PRETTYPRINT_SYMBOL_TABLE 10.2.1 control whether the symbol table should be included in the output.
Produce a html version of the symbol table, it’s module-independent, it’ll produce the same output for each module (the symbol table is global/unique).
HTML_PRETTYPRINT_SYMBOL_TABLE FALSE
Display the code without any decoration.
Display the code statements decorated with their transformers, except for loops, which are decorated with the transformer from the loop entering states to the loop body states. The effective loop transformer, linking the input to the output state of a loop, is recomputed when needed and can be deduced from the precondition of the next statement after the loop1 .
Display the code decorated with the complexities.
Display the code decorated with the preconditions.
Display the code decorated with the total preconditions.
Display the code decorated with the continuation preconditions.
Display the code decorated with the pointer regions.
Display the code decorated with the proper pointer regions.
Display the code decorated with the invariant read/write pointer regions.
Display the code decorated with the regions.
Display the code decorated with the proper regions.
Display the code decorated with the invariant read/write regions.
Display the code decorated with the IN regions.
Display the code decorated with the OUT regions.
Display the code decorated with the privatized regions.
Display the code decorated with complementary sections.
Display the code decorated with the proper pointer effects.
Display the code decorated with the proper effects.
Display the code decorated with the proper references.
Display the code decorated with the cumulated effects.
Display the code decorated with the cumulated effects.
Display the code decorated with the cumulated references.
Display the code decorated with its IN effects.
Display the code decorated with its OUT effects.
Display the code decorated with the cumulated effects.
Display the code decorated with the cumulated effects.
Display the code decorated with the proper reductions.
Display the code decorated with the cumulated reductions.
Display the code decorated with the static control.
Display the code decorated with the points to information.
Displays the code with simple pointer values relationships.
Displays the code with simple gen pointer values and kill sets.
PIPS can handle many different languages. By default the PrettyPrinter uses the native language as an output but it is also possible to prettyprint Fortran code as C code. Possible values for the PRETTYPRINT_LANGUAGE property are: native F95 F77 C.
PRETTYPRINT_LANGUAGE "native"
When prettyprinting semantic information (preconditions, transformers and regions), add a line before and after each piece of information if set to TRUE. The resulting code is more readable, but is larger.
PRETTYPRINT_LOOSE TRUE
By default, each prettyprinted line of Fortran or C code is terminated by its statement number in columns 73-80, unless no significative statement number is available. This feature is used to trace the origin of statements after program transformations and parallelization steps.
This feature may be inconvenient for some compilers or because it generates large source files. It may be turned off.
Note that the statement number is equal to the line number in the function file, that is the source file obtained after PIPS preprocessing2 and filtering3 , and not the user file, which is the file submitted by the user and which may contain several functions.
Note also that some phases in pips may add new statement that are not present in the original file. In this case the number of the statement that requires such a transformation, is used for the added statement.
PRETTYPRINT_STATEMENT_NUMBER TRUE
Note: this default value is overriden to FALSE by activate_language() for C and Fortran 95.
The structured control structure is shown by using an indentation. The default value is 3.
PRETTYPRINT_INDENTATION 3
Some people prefer to use a space after a comma to separate items in lists such as declaration lists or parameter lists in order to improve readability. Other people would rather pack more information per line. The default option is chosen for readability.
PRETTYPRINT_LISTS_WITH_SPACES TRUE
Depending on the user goal, it may be better to isolate comments used to display results of PIPS analyses from the source code statement. This is the default option.
PRETTYPRINT_ANALYSES_WITH_LF TRUE
This feature only exists for the semantics analyses.
Parallel output style How to print, from a syntactic point of view, a parallel do loop. Possible values are: do doall f90 hpf cray craft cmf omp.
PRETTYPRINT_PARALLEL "do"
Default sequential output style How to print, from a syntactic point of view, a parallel do loop for a sequential code. Of course, by default, the sequential output is sequential by definition, so the default value is "do".
But we may interested to change this behaviour to display after an application of internalize_parallel_code 8.1.8 the parallel code that is hidden in the sequential code. Possible values are: do doall f90 hpf cray craft cmf omp.
By default, parallel information is displayed with am OpenMP flavor since it is widely used nowadays.
PRETTYPRINT_SEQUENTIAL_STYLE "omp"
Add statement effects as comments in output; not implemented (that way) yet.
PRETTYPRINT_EFFECTS FALSE
The next property, PRETTYPRINT_IO_EFFECTS 10.2.22.4, is used to control the computation of implicit statement IO effects and display them as comments in output. The implicit effects on the logical unit are simulated by a read/write action to an element of the array TOP-LEVEL:LUNS(), or to the whole array when the element is not known at compile time. This is the standard behavior for PIPS. Some phases, e.g. hpfc, may turn this option off, but it is much more risky than to filter out abstract effects. Furthermore, the filtering is better because it takes into account all abstract effects, not only IO effects on logical units. PIPS users should definitely not turn off this property as the semantic equivalence between the inout and the output program is no longer guaranteed.
PRETTYPRINT_IO_EFFECTS TRUE
To transform C source code properly, variable and type declarations as well as variable and type references must be tracked alhtough standard use and def information is restricted to memory loads and stores because the optimizations are performed at a lower level. Fortran 77 analyses do not need information about variable declarations and there is not possibility of type definition. So the added information about variable declarations and references may be pure noise. It is possible to get rid of it by setting this property to TRUE, which is its default value before August 2010. For C code, it is better to set it to FALSE. For the time being, the default value cannot depend on the code language.
PRETTYPRINT_MEMORY_EFFECTS_ONLY FALSE
Transform DOALL loops into sequential loops with an opposed increment to check validity of the parallelization on a sequential machine. This property is not implemented.
PRETTYPRINT_REVERSE_DOALL FALSE
It is possible to print statement transformers as comments in code. This property is not intended for PIPS users, but is used internally. Transformers can be prettyprinted by using activate and PRINT_CODE_TRANSFORMERS
PRETTYPRINT_TRANSFORMER FALSE
It is possible to print statement preconditions as comments in code. This property is not intended for PIPS users, but is used internally. Preconditions can be prettyprinted by using activate and PRINT_CODE_PRECONDITIONS
PRETTYPRINT_EXECUTION_CONTEXT FALSE
It is possible to print statement with convex array region information as comments in code. This property is not intended for PIPS users, but is used internally. Convex array regions can be prettyprinted by using activate and PRINT_CODE_REGIONS or PRINT_CODE_PROPER_REGIONS
PRETTYPRINT_REGION FALSE
By default, convex array regions are printed for arrays only, but the internal representation includes scalar variables as well. The default option can be overriden with this property.
PRETTYPRINT_SCALAR_REGIONS FALSE
All these debugging options should be set to FALSE for normal operation, when the prettyprinter is expected to produce code as close as possible to the input form. When they are turned on, the output is closer to the PIPS internal representation.
Sequences are implicit in Fortran and in many programming languages but they are internally represented. It is possible to print pieces of information gathered about sequences by turning on this property.
PRETTYPRINT_BLOCKS FALSE
To print all the C blocks (the { } in C, you can set the following property:
PRETTYPRINT_ALL_C_BLOCKS FALSE
This property is a C-specialized version of PRETTYPRINT_BLOCKS, since in C you can represent the blocks. You can combine this property with a PRETTYPRINT_EMPTY_BLOCKS set to true too. Right now, the prettyprint of the C block is done in the wrong way, so if you use this option, you will have redundant blocks inside instructions, but you will have all the other hidden blocks too...
To print unstructured statements:
PRETTYPRINT_UNSTRUCTURED FALSE
Print all effects for all statements regardless of PRETTYPRINT_BLOCKS 10.2.22.5 and PRETTYPRINT_UNSTRUCTURED 10.2.22.5.
PRETTYPRINT_ALL_EFFECTS FALSE
Print empty statement blocks (false by default):
PRETTYPRINT_EMPTY_BLOCKS FALSE
Print statement ordering information (false by default):
PRETTYPRINT_STATEMENT_ORDERING FALSE
The next property controls the print out of DO loops and CONTINUE statement. The code may be prettyprinted with DO label and CONTINUE instead of DO-ENDDO, as well as with other useless CONTINUE (This property encompasses a virtual PRETTYPRINT_ALL_CONTINUE_STATEMENTS). If set to FALSE, the default option, all useless CONTINUE statements are NOT prettyprinted (ie. all those in structured parts of the code). This mostly is a debugging option useful to understand better what is in the internal representation.
Warning: if set to TRUE, generated code may be wrong after some code transformations like distribution...
PRETTYPRINT_ALL_LABELS FALSE
Print code with DO label as comment.
PRETTYPRINT_DO_LABEL_AS_COMMENT FALSE
Print private variables without regard for their effective use. By default, private variables are shown only for parallel DO loops.
PRETTYPRINT_ALL_PRIVATE_VARIABLES FALSE
Non-standard variables and tests are generated to simulate the control effect of Fortran IO statements. If an end-of-file condition is encountered or if an io-error is raised, a jump to relevant labels may occur if clauses ERR= or END= are defined in the IO control list. These tests are normally not printed because they could not be compiled by a standard Fortran compiler and because they are redundant with the IO statement itself.
PRETTYPRINT_CHECK_IO_STATEMENTS FALSE
Print the final RETURN statement, although this is useless according to Fortran standard. Note that comments attached to the final return are lost if it is not printed. Note also that the final RETURN may be part of an unstructured in which case the previous property is required.
PRETTYPRINT_FINAL_RETURN FALSE
The internal representation is based on a standard IF structure, known as block if in Fortran jargon. When possible, the prettyprinter uses the logical if syntactical form to save lines and to produce an output assumed closer to the input. When statements are decorated, information gathered by PIPS may be lost. This property can be turned on to have an output closer to the internal representation. Note that edges of the control flow graphs may still be displayed as logical if since they never carry any useful information4 .
PRETTYPRINT_BLOCK_IF_ONLY FALSE
Effects give data that may be read and written in a procedure. These data are represented by their entity name. By default the entity name used is the shortest nom-ambiguous one. The PRETTYPRINT_EFFECT_WITH_FULL_ENTITY_NAME 10.2.22.5.1 property can be used to force the usage of full entity name (module name + scope + local name).
PRETTYPRINT_EFFECT_WITH_FULL_ENTITY_NAME FALSE
In order to have information on the scope of commons, we need to know the common in which the entity is declared if any. To get this information the PRETTYPRINT_WITH_COMMON_NAMES 10.2.22.5.1 property has to set to TRUE.
PRETTYPRINT_WITH_COMMON_NAMES FALSE
By default, expressions are simplified according to operator precedences. It is possible to override this prettyprinting option and to reflect the abstract tree with redundant parentheses.
PRETTYPRINT_ALL_PARENTHESES FALSE
By default, the C prettyprinter uses a minimum of braces to improve readability. However, gcc advocates the use of more braces to avoid some ambiguities about else clauses. In order to run succesfully gcc with options -Wall and -Werror, it is possible to force the print-out of all possible braces.
PRETTYPRINT_ALL_C_BRACES FALSE
The previous property leads to hard to read source code. Property PRETTYPRINT_GCC_C_BRACES 10.2.22.5.1 is used to print only a few additional braces required by gcc to avoid ambiguous else warning messages.
PRETTYPRINT_GCC_C_BRACES FALSE
By default in Fortran (and not in C), module declarations are preserved as huge strings to produce an output as close as possible to the input (see field decls_text in type code). However, large program transformations and code generation phases, e.g. hpfc, require updated declarations.
Regenerate all variable declarations, including those variables not declared in the user program. By default in Fortran, when possible, the user declaration text is used to preserve comments.
PRETTYPRINT_ALL_DECLARATIONS FALSE
If the prettyprint of the header and the declarations are done by PIPS, try to display the genuine comments. Unfortunately, there is no longer order relation between the comments and the declarations since these are sorted by PIPS. By default, do not try to display the comments when PIPS is generating the header.
PRETTYPRINT_HEADER_COMMENTS FALSE
How to regenerate the common declarations. It can be none, declaration, or include.
PRETTYPRINT_COMMONS "declaration"
DATA declarations are partially handled presently.
PRETTYPRINT_DATA_STATEMENTS TRUE
Where to put the dimension information, which must appear once. The default is associated to the type information. It can be associated to The type, or preferably to the common if any, or maybe to a dimension statement, which is not implemented.
PRETTYPRINT_VARIABLE_DIMENSIONS "type"
Print transformers, preconditions and regions in a format accepted by Foresys and Partita. Not maintained.
PRETTYPRINT_FOR_FORESYS FALSE
To deal specifically with the prettyprint for hpfc
PRETTYPRINT_HPFC FALSE
The following property tells PIPS to attach various Emacs properties for interactive purpose. Used internally by the Emacs pretyyprinter and the epip user interface.
PRETTYPRINT_ADD_EMACS_PROPERTIES FALSE
These are files containing a pretty-printed version of code to be displayed with its intraprocedural control graph as a graph, for example using the uDrawGraph5 program (formerly known as daVinci) or dot/GraphViz tools. More concretely, use some scripts like pips_unstructured2daVinci or pips_unstructured2dot to display graphically these .pref-graph files.
The statements may be decorated with complexities, preconditions, transformers, regions,… depending on the printer used to produce this file.
Display the code without any decoration.
Display the code decorated with the transformers.
Display the code decorated with the complexities.
Display the code decorated with the preconditions.
Display the code decorated with the preconditions.
Display the code decorated with the regions.
Display the code decorated with the IN regions.
Display the code decorated with the OUT regions.
Display the code decorated with the proper effects.
Display the code decorated with the cumulated effects.
This prettyprinter is NOT a call graph prettyprinter (see Section 6.1). Control flow information can be displayed and every call site is shown, possibly with some annotation like precondition or region
This prettyprinter uses the module codes in the workspace database to build the ICFG.
Print IF statements controlling call sites:
ICFG_IFs FALSE
Print DO loops enclosing call sites:
ICFG_DOs FALSE
It is possible to print the interprocedural control flow graph as text or as a graph using daVinci format. By default, the text output is selected.
ICFG_DV FALSE
To be destroyed:
ICFG_CALLEES_TOPO_SORT FALSE
ICFG_DECOR 0
ICFG_DRAW TRUE
ICFG default indentation when going into a function or a structure.
ICFG_INDENTATION 4
Debugging level (should be ICFG_DEBUG_LEVEL and numeric instead of boolean!):
ICFG_DEBUG FALSE
Effects are often much too numerous to produce a useful interprocedural control flow graph.
The integer property RW_FILTERED_EFFECTS 10.3.12 is used to specify a filtering criterion.
RW_FILTERED_EFFECTS 0
To output a code with a hierarchical view of the control graph with markers instead of a flat one. It purposes a display with a graph browser such as daVinci6 :
PRETTYPRINT_UNSTRUCTURED_AS_A_GRAPH FALSE
and to have a decorated output with the hexadecimal addresses of the control nodes:
PRETTYPRINT_UNSTRUCTURED_AS_A_GRAPH_VERBOSE FALSE
File containing a pretty-printed version of a parallelized_code. Several versions are available. The first one is based on Fortran-77, extended with a DOALL construct. The second one is based on Fortran-90. The third one generates CRAY Research directives as comments if and only if the correspondent parallelization option was selected (see Sectionsubsection-parallelization).
No one knows why there is no underscore between parallel and printed...
Output a Fortran-77 code extended with DOALL parallel constructs.
Output the code decorated with HPF directives.
Output the code decorated with OpenMP (OMP) directives.
Output the code with some Fortran-90 array construct style.
Output the code decorated with parallel Cray directives. Note that the Cray parallelization algorithm should have been used in order to match Cray directives for parallel vector processors.
This kind of file contains the sub call graph7 of a module. Of course, the call graph associated to the MAIN module is the program call graph.
Each module can be decorated by summary information computed by one of PIPS analyses.
If one module has different callers, its sub call tree is replicated once for each caller8 .
No fun to read, but how could we avoid it with a text output? But it is useful to check large analyses.
The resource defined in this section is callgraph_file (note the missing underscore between call and graph in callgraph...). This is a file resource to be displayed, which cannot be loaded in memory by pipsdbm.
Note that the input resource lists could be reduced to one resource, the decoration. pipsmake would deduce the other ones. There is no need for a transitive closure, but some people like it that way to make resource usage verification possible...RK: explain... FI: no idea; we would like to display any set of resources, but the sets are too numerous to have a phase for each.
Aliases for call graphs must de different from aliases for interprocedural control flow graphs (ICFG). A simple trick, a trailing SPACE character, is used.
To have the call graph without any decoration.
To have the call graph decorated with the complexities.
To have the call graph decorated with the preconditions.
To have the call graph decorated with the total preconditions.
To have the call graph decorated with the transformers.
To have the call graph decorated with the proper effects.
To have the call graph decorated with the cumulated effects.
To have the call graph decorated with the regions.
To have the call graph decorated with the IN regions.
To have the call graph decorated with the OUT regions.
This library is used to display the calling relationship between modules. It is different from the interprocedural call flow graph, ICFG (see Section 10.3.12). For example: if A calls B twice, in callgraph, there is only one edge between A and B; while in ICFG (see next section)), there are two edges between A and B, since A contains two call sites.
The call graph is derived from the modules declarations. It does not really the parsed code per se, but the code must have been parsed to have up-to-date declarations in the symbol table.
Because of printout limitations, the call graph is developed into a tree before it is printed. The sub-graph of a module appears as many times as is has callers. The resulting printout may be very long.
There is no option for the callgraph prettyprinter except for debugging.
Debugging level (should be CALLGRAPH_DEBUG_LEVEL and numeric!)
CALLGRAPH_DEBUG FALSE
This is the file ICFG with format of graph uDrawGraph9 (formerly daVinci). This should be generalized to be less tool-dependent.
Display the ICFG graphically decorated with the write proper effects filtered for a variable.
This kind of file contains a more or less precise interprocedural control graph. The graph can be restricted to call sites only, to call sites and enclosing DO loops or to call sites, enclosing DO loops and controlling IF tests. This abstraction option is orthogonal to the set of decorations, but pipsmake does not support this orthogonality. All combinations are listed below.
Each call site can be decorated by associated information computed by one of PIPS analyses.
Note: In order to avoid conflicts with callgraph aliases, a space character is appended at each alias shared with call graph related functions (Guillaume Oget).
Display the plain ICFG, without any decoration.
Display the ICFG decorated with complexities.
Display the ICFG decorated with preconditions. They are expressed in the callee name space to evaluate the interest of cloning, depending on the information available to the callee at a given call site.
Display the ICFG decorated with total preconditions. They are expressed in the callee name space to evaluate the interest of cloning, depending on the information available to the callee at a given call site.
Display the ICFG decorated with transformers.
Display the ICFG decorated with the proper effects.
Display the ICFG decorated with the write proper effects filtered for a variable.
Display the ICFG decorated with cumulated effects.
Display the ICFG decorated with regions.
Display the ICFG decorated with IN regions.
Display the ICFG decorated with OUT regions.
Display the plain ICFG with loops, without any decoration.
Display the ICFG decorated with loops and complexities.
Display the ICFG decorated with preconditions.
Display the ICFG decorated with total preconditions.
Display the ICFG decorated with transformers.
Display the ICFG decorated with proper effects.
Display the ICFG decorated with cumulated effects.
Display the ICFG decorated with regions.
Display the ICFG decorated with IN regions.
Display the ICFG decorated with the OUT regions.
Display the plain ICFG with loops, without any decoration.
Display the ICFG decorated with the complexities.
Display the ICFG decorated with the preconditions.
Display the ICFG decorated with the preconditions.
Display the ICFG decorated with the transformers.
Display the ICFG decorated with the proper effects.
Display the ICFG decorated with the cumulated effects.
Display the ICFG decorated with the regions.
Display the ICFG decorated with the IN regions.
Display the ICFG decorated with the OUT regions.
This file shows the dependence graph.
Known bug: there is no precise relationship between the dependence graph seen by the parallelization algorithm selected and any of its view...
Two formats are available: the default format which includes dependence cone and a SRU format which packs all information about one arc on one line and which replaces the dependence cone by the dependence direction vector (DDV). The line numbers given with this format are in fact relative (approximatively...) to the statement line in the PIPS output. The SRU format was defined with researchers at Slippery Rock University (PA). The property
PRINT_DEPENDENCE_GRAPH_USING_SRU_FORMAT 6.6.6.4
is set to FALSE by default.
Display dependence levels for loop-carried and non-loop-carried dependence arcs due to non-privatized variables. Do not display dependence cones.
Display dependence levels for loop-carried dependence arcs only. Ignore arcs labeled by private variables and do not print dependence cones.
Display dependence levels and dependence polyhedra/cones for all dependence arcs, whether they are loop carried or not, whether they are due to a private variable (and ignored by parallelization algorithms) or not. Dependence cones labeling arcs are printed too.
Same as print_whole_dependence_graph 10.8.4 but it’s filtered by some variables. Variables to filter is a comma separated list set by user via property “EFFECTS_FILTER_ON_VARIABLE”.
Same as print_filtered_dependence_graph 10.8.5 but its output is uDrawGraph10 format.
Check impact of alias on the dependance graph.RK: What is that? FI: Maybe, we should sign how contributions? See validation?
Display the chains graph in a textual format. The following properties control the output :
PRINT_DEPENDENCE_GRAPH 6.6.6.4 PRINT_DEPENDENCE_GRAPH_WITHOUT_PRIVATIZED_DEPS 6.6.6.4 PRINT_DEPENDENCE_GRAPH_WITHOUT_NOLOOPCARRIED_DEPS 6.6.6.4 PRINT_DEPENDENCE_GRAPH_WITH_DEPENDENCE_CONES 6.6.6.4 PRINT_DEPENDENCE_GRAPH_USING_SRU_FORMAT 6.6.6.4
Display the chains graph in graphviz dot format.
Display the dependence graph in graphviz dot format.
Using pyps, some convenient functions are provide, for instance :
which will produce in current directory a file named my_function_name.png.
Here are the properties available to tune the Dot output, read dot documentation for available colors, style, shape, etc.
PRINT_DOTDG_STATEMENT TRUE
Print statement code and not only ordering inside nodes.
PRINT_DOTDG_TOP_DOWN_ORDERED TRUE
Add a constraint on top-down ordering for node instead of free dot placement.
PRINT_DOTDG_CENTERED FALSE
Should dot produce a centered graph ?
PRINT_DOTDG_TITLE ""
PRINT_DOTDG_TITLE_POSITION "b"
Title and title position (t for top and b for bottom) for the graph.
PRINT_DOTDG_BACKGROUND "white"
Main Background.
PRINT_DOTDG_NODE_SHAPE "box"
Shape for statement nodes.
PRINT_DOTDG_NODE_SHAPE_COLOR "black"
PRINT_DOTDG_NODE_FILL_COLOR "white"
PRINT_DOTDG_NODE_FONT_COLOR "black"
PRINT_DOTDG_NODE_FONT_SIZE "18"
PRINT_DOTDG_NODE_FONT_FACE "Times-Roman"
Color for the shape, background, and font of each node.
PRINT_DOTDG_FLOW_DEP_COLOR "red"
PRINT_DOTDG_ANTI_DEP_COLOR "green"
PRINT_DOTDG_OUTPUT_DEP_COLOR "blue"
PRINT_DOTDG_INPUT_DEP_COLOR "black"
Color for each type of dependence
PRINT_DOTDG_FLOW_DEP_STYLE "solid"
PRINT_DOTDG_ANTI_DEP_STYLE "solid"
PRINT_DOTDG_OUTPUT_DEP_STYLE "solid"
PRINT_DOTDG_INPUT_DEP_STYLE "dashed"
Style for each type of dependence
A basic and experimental C dumper to output a Fortran program in C code. It is not the same as the default pretty printer that is normaly used to pretty print C code in C. This pass is mainly use inside PIPS and Par4All to be able to generate call to CUDA kernels from a fortran code. The kernel is supposed not to use I/O intrinsics (such as WRITE, READ), they are not handle at the moment (and not usefull in the CUDA context) by the crough printer. However it is still possible to print the C code with the name of the fortran intrinsic using the property CROUGH_PRINT_UNKNOWN_INTRINSIC.
Display the crough output of a fortran function.
Once C version of fortran code has been generated, ones might like to call this C functions from fortran code. A convenient way to do this is to use an interface in the fortran code. PIPS can generate the interface module for any function using the following pass.
By default the crough pass fails if it encounters a fortran intrinsic that cannot be translated to C. However it is still possible to print the C code with the name of the fortran intrinsic using the following property.
CROUGH_PRINT_UNKNOWN_INTRINSIC FALSE
By default the crough pass tries to match the best C type for any Fortran type. Here is the matches between Fortran and C variables:
However, many Fortran compilers (ifort, gfortran) allow to change the type size at compile time. It is possible to do the same by setting the property CROUGH_USER_DEFINED_TYPE to TRUE. In such a case the include file specified by the property CROUGH_INCLUDE_FILE is included by any file generated using crough. It has to define (using #define or typedef) the two types defined by the properties CROUGH_INTEGER_TYPE and CROUGH_REAL_TYPE. Obviously those types are used in the generated C file to declare variables that has the types INTEGER or REAL in the original Fortan file. When choosing that solution all INTEGER and REAL (including INTEGER*4, INTEGER*8, REAL*4 and REAL*8) variables will be set to the same user defined types.
CROUGH_USER_DEFINED_TYPE FALSE
CROUGH_INCLUDE_FILE "p4a_crough_types.h"
CROUGH_INTEGER_TYPE "p4a_int"
CROUGH_REAL_TYPE "p4a_real"
Is is possible to prettyprint function parameters that are arrays as pointers using the property CROUGH_ARRAY_PARAMETER_AS_POINTER
CROUGH_ARRAY_PARAMETER_AS_POINTER FALSE
When PRETTYPRINT_C_FUNCTION_NAME_WITH_UNDERSCORE is set to TRUE, an underscore is added at the end of the module name. This is needed when translating only some part of a Fortran Program to C. This property must be used with great care, so that only interface function names are changed : the function names in subsequent calls are not modified. An other solution to call a C function from a fortran program is to use/declare an interface in the fortran source code (This feature is part of the Fortran 2003 standard but many Fortran95 compilers support it). The property CROUGH_FORTRAN_USES_INTERFACE can be set to TRUE when the Fotran code integrates interfaces. In such a case, the unmodified scalar function parameters (by the function or any of its callees) are expected to be passed by values, the other parameters are passed by pointers. Finally when using interfaces it is also possible to pass all the scalar variables by values using the propety CROUGH_SCALAR_BY_VALUE_IN_FCT_DECL. Now, let’s talk about function called from the fortran code PIPS has to print in C. The problem on how to pass scalars (by value or by pointer) also exists. At the moment PIPS is less flexible for function call. One of the solution has to be choosen using the property CROUGH_SCALAR_BY_VALUE_IN_FCT_CALL.
PRETTYPRINT_C_FUNCTION_NAME_WITH_UNDERSCORE FALSE
CROUGH_FORTRAN_USES_INTERFACE FALSE
CROUGH_SCALAR_BY_VALUE_IN_FCT_DECL FALSE
CROUGH_SCALAR_BY_VALUE_IN_FCT_CALL FALSE
If the property DO_RETURN_TYPE_AS_TYPEDEF is set to TRUE the crough phase additionaly does the same thing that the phase set_return_type_as_typedef does (cf section 2). In such a case the same property SET_RETURN_TYPE_AS_TYPEDEF_NEW_TYPE is taken into account. This is only posible for the SUBROUTINES and the FUNCTIONS but not for the PROGRAMS
DO_RETURN_TYPE_AS_TYPEDEF FALSE
Using the property INCLUDE_FILES_LIST, it is possible to insert some #include statement before to output the code. The INCLUDE_FILES_LIST is a string interpreted as coma (and/or blank) separated list of files.
CROUGH_INCLUDE_FILE_LIST ""
This pass is used by the phrase project, which is an attempt to automatically (or semi-automatically) transform high-level language application into control code with reconfigurable logic accelerators (such as fpgas or data-paths with alu).
This pass is used in context of PHRASE project for synthetisation of reconfigurable logic for a portion of initial code. This function can be viewed as a SmallTalk pretty-printer of a subset of Fortran or C.
It is used as input for the Madeo synthesis tools from UBO/AS that is written in SmallTalk and take circuit behaviour in SmallTalk.
It is an interesting language fusion...
This pass is used for printing pragmas scop and endscop which delimit the static control parts (SCoP) of the code. Instrumented code can be an input for any PoCC compiler.
For the outlining of the control static parts, function prefix names to be used during the generation:
SCOP_PREFIX "SCoP"
Because of user function calls are not allowed in PoCC static control parts, whereas they are generally accepted in PIPS (see Arnauld Leservot’s PhD), we introduce a property to control the impact of user calls. This property could be called POCC_COMPATIBILIRY if there were a set of limiting rules to apply for PoCC, but user calls are the only current issue11 So the property is called STATIC_CONTROLIZE_ACROSS_USER_CALLS 10.11 and its default value is TRUE.
STATIC_CONTROLIZE_ACROSS_USER_CALLS TRUE
This pass is used for the DREAM-UP project. The internal representation of a C or Fortran program is dumped as CLAIRE objects, either DATA_ARRAY or TASK. CLAIRE is an object-oriented language used to develop constraint solvers.
The only type constructor is array. Basic types must be storable on a fixed number of bytes.
The code structure must be a sequence of loop nests. Loops must be perfectly nested and parallel. Each loop body must be a single array assignment. The right-hand side expression must be a function call.
If the input code does not meet these conditions, a user error is generated.
This pass is used for the specification input and transformation in the XML format which can further be used by number of application as input. This function can be viewed as a XML pretty-printer of a subset of C and Fortran programs.
This phase was developped for the DREAM-UP/Ter@ops project to generate models of functions used for automatic mapping by APOTRES []. It generates XML code like the print_xml_code pass, but the input contains explicitly loops to scan motifs. It is useless for other purposes. RK: gnih? FI: to be deleted? CA: more to say?
This pass is used in the DREAM-UP project for module specification input and transformation (?) []. This function can be viewed as a CLAIRE pretty-printer of a subset of Fortran.
This pass generates CLAIRE code like the print_claire_code pass, but the input contains explicitly loops to scan motifs.
This pass was developped for the Ter@ops project to generate models of functions and application used for automatic mapping by SPEAR. It generates XML code.
This part of PIPS was implemented at Centre d’ï¿œtudes Atomiques, Limeil-Brï¿œvannes, by Benoï¿œt de Dinechin, Arnauld Leservot and Alexis Platonoff.
Unfortunately, this part is no longer used in PIPS right now because of some typing issues in the code. To be fixed when somebody needs it.
static_controlize 11.1 transforms all the loops in order to have steps equal to one. Only loops with constant step different than one are normalized. Normalized loop counters are instantiated as a new kind of entity: NLC. This entity is forwarded in the inner statements. It also gets the structural parameters and makes new ones when it is possible (“NSP”). It detects enclosing loops, enclosing tests and the static_control property for each statement. Those three informations are mapped on statements. Function static_controlize 11.1 also modifies code (> MODULE.code). It is not specified here for implementation bug purpose.
The definition of a static control program is given in [?].
See the alias print_code_static_control 10.2.19 and function print_code_static_control 10.2.19 in Section 10.1 and so on.
Function scheduling computes a schedule, called Base De Temps in French, for each assignment instruction of the program. This computation is based on the Array Data Flow Graph (see [?, ?]).
The output of the scheduling is of the following form: (the statements are named in the same manner as in the array DFG)
Function reindexing transforms the code using the schedule (bdt) and the mapping (plc) (see [?, ?]). The result is a new resource named reindexed_code.
How to get a pretty-printed version of reindexed_code ? Two prettyprinters are available. The first one produces CM Fortran and the result is stored in a file suffixed by .fcm. The second one produces CRAFT Fortran and the result is stored in a file suffixed by .craft.
Use the polyhedric method to parallelize the code and display the reindexed code in a CMF (parallel Fortran extension from TMC, Thinking Machine Corporation) style.
Use the polyhedric method to parallelize the code and display the reindexed code in a CRAFT (parallel Fortran used on the Cray T3 serie) style.
For presentation issues, it is useful to select only the features that are needed by a user and to display them in a comprehensive order. For that purpose, a layout description mechanism is used here to pick among the PIPS phases described above.
For each menu, the left part before the arrow, ->, is the menu item title and the right part is the PIPS procedure to be called when the item is selected. For the view menu (section 12.1, there is two display methods to view resources separated by a comma, the first one is the method for wpips, the second one is the one used in epip, followed by the icon to use.
Use a blank line to insert a menu separator.
The view menu is displayed according to the following layout and methods (wpips method, epip method, icon name for the frame):
View
The transformation menu is displayed as here:
Transformations
At the end of this menu is added a special entry in wpips, the “Edit” line that allows the user to edit the original file. It is seen as a very special transformation, since the user can apply whatever transformation (s)he wants...
New functionalities can easily be added to PIPS. The new pass names must be declared somewhere in this file as well as the resources required and produced. Then, make must be run in the Documentation directory and the pipsmake library must be recompiled and PIPS interfaces (pips, tpips, wpips) linked with the new C modules.
It is much more difficult to add a new type of resources, because PIPS database manager, pipsdbm, is not as automatized as pipsmake. This is explained in [?].
Abort, 8
Abstract Syntax Tree, 18
Alias Analysis, 76
Alias Checking, 87
Alias Classes, 76
Alias Propagation, 86
Allen & Kennedy Algorithm, 91
Allen&Kennedy, 90
Alternate Return, 22
Analysis, 34, 188
Analysis (Semantics), 51
Array access, 84
Array Expansion, 171, 172
Array Privatization, 168, 170
Array Region, 70, 75, 83
ARRAY_PRIV_FALSE_DEP_ONLY, 170
ARRAY_SECTION_PRIV_COPY_OUT, 170
Assigned GO TO, 24
AST, 18
Atomic Chains, 43
Atomization, 142
ATOMIZE_INDIRECT_REF_ONLY, 142
Atomizer, 94, 141
atomizer, 141
Automatic Distribution, 100
Buffer overflow, 84
C3 Linear Library, 7
Call Graph, 34, 199
Callees, 19
CFG, 28
CHAINS_DATAFLOW_DEPENDENCE_ONLY, 45
CHAINS_MASK_EFFECTS, 45
checkpoint, 7
CLEAN_UP_SEQUENCES_DISPLAY_STATISTICS, 30
Cloning, 158
CM Fortran, 189
Code Distribution, 106
Code Prettyprinter, 188
Common subexpression elimination, 142, 145, 149
compilation unit, 25
Complementary Sections, 83
Complementary Sections (Summary), 83
Complex Constant, 13
Complexity, 66, 67, 179
Complexity (Floating Point), 67
Complexity (Summary), 67
Complexity (Uniform), 67
COMPLEXITY_COST_TABLE, 68
COMPLEXITY_EARLY_EVALUATION, 69
COMPLEXITY_INTERMEDIATES, 68
COMPLEXITY_PARAMETERS, 68
COMPLEXITY_PRINT_COST_TABLE, 68
COMPLEXITY_PRINT_STATISTICS, 68
COMPLEXITY_TRACE_CALLS, 68
COMPUTE_ALL_DEPENDENCES, 50
Computed GO TO, 24
Control Counters, 141
Control Flow Graph, 28
Control Restructurer, 134, 138
Control Simplification, 131
Controlizer, 28
correctness, 21
Craft, 189
Cray, 91
Cray Fortran, 189
CSE, 142, 149
Cumulated Effects, 36
DaVinci, 198
Dead Code, 132
Dead Code Elimination, 131
Dead code elimination, 133
DEAD_CODE_DISPLAY_STATISTICS, 132
Debug, 191
Debug (Complexity), 68
Debug (Semantics), 65
Debugging, 7
Declaration, 193
Def-Use Chains, 42, 49
Dependence Graph, 45, 50, 212
Dependence Test, 47
dependence test
fast, 46
full, 47
regions, 47
semantics, 47
Dependence test statistics, 48
DEPENDENCE_TEST, 47
DESTRUCTURE_FORLOOPS, 137
DESTRUCTURE_LOOPS, 137
DESTRUCTURE_TESTS, 137
DESTRUCTURE_WHILELOOPS, 137
DG, 45, 212
DG Prettyprinter, 50
DISJUNCT_IN_OUT_REGIONS, 75
DISJUNCT_REGIONS, 75
Distribution, 89, 100
Distribution (Loop), 122
Distribution init, 107
DREAM-UP, 219
Dynamic Aliases, 76
Effect, 35, 40
Effects (Cumulated), 36
Effects (IN), 37
Effects (Memory), 38
Effects (OUT), 37
Effects (Proper), 35
EFFECTS_PRINT_SDFI, 38
Emacs, 194
Emulated Shared Memory, 100
Entity, 18
ENTRY, 22
EXACT_REGIONS, 75
Expansion, 171
Expression, 62
Final Postcondition, 60
Finite State Machine Generation, 139
Fix Point, 63
Floating Point Complexity, 67
Flow Sensitivity, 61
Foresys, 194
Format (Fortran), 30
Fortran (Cray), 189
Fortran 90, 21, 189
Forward substitution, 145
freeze variables, 172
FSM Generation, 140
FSMIZE_WITH_GLOBAL_VARIABLE, 140
FUSE_CONTROL_NODES_WITH_COMMENTS_OR_LABEL, 30
GATHER_FORMATS_AT_BEGINNING, 30
GATHER_FORMATS_AT_END, 30
General Loop Interchange, 127
GENERATE_NESTED_PARALLEL_LOOPS, 90
GLOBAL_EFFECTS_TRANSLATION, 103
GO TO (Assigned), 24
GO TO (Computed), 24
hCFG, 28
Hierarchical Control Flow Graph, 28
Hollerith, 12
HPF, 101, 103, 189, 194
HPFC, 101
HPFC_BUFFER_SIZE, 103
HPFC_DYNAMIC_LIVENESS, 103
HPFC_EXPAND_CMPLID, 103
HPFC_EXPAND_COMPUTE_COMPUTER, 103
HPFC_EXPAND_COMPUTE_LOCAL_INDEX, 103
HPFC_EXPAND_COMPUTE_OWNER, 103
HPFC_EXTRACT_EQUALITIES, 103
HPFC_EXTRACT_LATTICE, 103
HPFC_FILTER_CALLEES, 103
HPFC_GUARDED_TWINS, 103
HPFC_IGNORE_FCD_SET, 103
HPFC_IGNORE_FCD_SYNCHRO, 103
HPFC_IGNORE_FCD_TIME, 103
HPFC_IGNORE_IN_OUT_REGIONS, 103
HPFC_IGNORE_MAY_IN_IO, 103
HPFC_LAZY_MESSAGES, 103
HPFC_NO_WARNING, 103
HPFC_OPTIMIZE_REMAPPINGS, 103
HPFC_REDUNDANT_SYSTEMS_FOR_REMAPS, 103
HPFC_SYNCHRONIZE_IO, 103
HPFC_TIME_REMAPPINGS, 103
HPFC_USE_BUFFERS, 103
Hyperplane Method, 128
ICFG, 197
ICFG_CALLEES_TOPO_SORT, 197
ICFG_DEBUG, 197
ICFG_DECOR, 197
ICFG_DOs, 197
ICFG_DRAW, 197
ICFG_DV, 197
ICFG_IFs, 197
ICFG_INDENTATION, 197
If Conversion, 97
If Simplification, 131
Implicit None, 13
IN Effects, 37
IN Regions, 73
IN Summary Regions, 74
Include, 12, 13
Index Set Splitting, 125
Initial Precondition, 55
Inlining, 154
INLINING_CALLERS, 154
Input File, 11
Interprocedural, 62
Interprocedural Points to Analysis, 77, 78
Intraprocedural Points-to Analysis, 77
Intraprocedural Summary Precondition, 55
Invariant code motion, 145
invariant code motion, 130
IR, 18
Kaapi, 107
KEEP_READ_READ_DEPENDENCE, 45
Logging, 6, 8
Loop Distribution, 122
Loop fusion, 124
Loop Interchange, 127
Loop Normalize, 129
Loop Simplification, 131
Loop Tiling, 95
Loop Unroll, 95
Loop Unrolling, 125
LOOP_LABEL, 121
MAY Region, 71
Memory Effect, 35
Memory Effects, 38
MEMORY_EFFECTS_ONLY, 38
Missing Code, 15
Missing file, 12
Module, 18
MPI, 105
MUST Region, 72
MUST_REGIONS, 75
NewGen, 7
NO_USER_WARNING, 10
OpenGPU, 218
OpenMP, 105, 174
OUT Effects, 37
OUT Regions, 74
OUT Summary Regions, 74
Parallelization, 89–91
PARALLELIZATION_STATISTICS, 90
Parsed Code, 19
PARSER_ACCEPT_ANSI_EXTENSIONS, 21
PARSER_ACCEPT_ARRAY_RANGE_EXTENSION, 21
PARSER_EXPAND_STATEMENT_FUNCTIONS, 24
PARSER_FORMAL_LABEL_SUBSTITUTE_PREFIX, 22
PARSER_LINEARIZE_LOOP_BOUNDS, 22
PARSER_RETURN_CODE_VARIABLE, 22
PARSER_SIMPLIFY_LABELLED_LOOPS, 22
PARSER_SUBSTITUTE_ALTERNATE_RETURNS, 22
PARSER_SUBSTITUTE_ASSIGNED_GOTO, 24
PARSER_SUBSTITUTE_ENTRIES, 22
PARSER_TYPE_CHECK_CALL_SITES, 21
PARSER_WARN_FOR_COLUMNS_73_80, 20
Partial Evaluation, 143
PARTIAL_DISTRIBUTION, 122
PHRASE, 106, 139, 218
Phrase comEngine Distributor, 108
Phrase Distributor, 107
Phrase Distributor Control Code, 107
Phrase Distributor Initialisation, 106
Phrase Remove Dependences, 108
Pipsdbm, 8
PIPSDBM_NO_FREE_ON_QUIT, 8
Pipsmake, 7
Pointer Values Analyses, 78
Postcondition (Final), 60
Precondition, 54, 60
Precondition (Initial), 55
Precondition (Summary), 55, 56
Preprocessing, 12, 13
PRETTYPRINT_ADD_EMACS_PROPERTIES, 194
PRETTYPRINT_ALL_C_BLOCKS, 191
PRETTYPRINT_ALL_DECLARATIONS, 193
PRETTYPRINT_ALL_EFFECTS, 191
PRETTYPRINT_ALL_LABELS, 191
PRETTYPRINT_ALL_PARENTHESES, 191
PRETTYPRINT_ALL_PRIVATE_VARIABLES, 191
PRETTYPRINT_ANALYSES_WITH_LF, 188
PRETTYPRINT_BLOCK_IF_ONLY, 191
PRETTYPRINT_BLOCKS, 191
PRETTYPRINT_C_CODE, 188
PRETTYPRINT_CHECK_IO_STATEMENTS, 191
PRETTYPRINT_COMMONS, 193
PRETTYPRINT_DO_LABEL_AS_COMMENT, 191
PRETTYPRINT_EFFECTS, 190
PRETTYPRINT_EMPTY_BLOCKS, 191
PRETTYPRINT_EXECUTION_CONTEXT, 190
PRETTYPRINT_FINAL_RETURN, 191
PRETTYPRINT_FOR_FORESYS, 194
PRETTYPRINT_HEADER_COMMENTS, 193
PRETTYPRINT_HPFC, 194
PRETTYPRINT_INDENTATION, 188
PRETTYPRINT_INTERNAL_RETURN, 191
PRETTYPRINT_IO_EFFECTS, 190
PRETTYPRINT_LISTS_WITH_SPACES, 188
PRETTYPRINT_LOOSE, 188
PRETTYPRINT_MEMORY_EFFECTS_ONLY, 190
PRETTYPRINT_PARALLEL, 189
PRETTYPRINT_REGENERATE_ALTERNATE_RETURNS, 22
PRETTYPRINT_REGION, 190
PRETTYPRINT_REVERSE_DOALL, 190
PRETTYPRINT_SCALAR_REGIONS, 190
PRETTYPRINT_STATEMENT_NUMBER, 188
PRETTYPRINT_STATEMENT_ORDERING, 191
PRETTYPRINT_TRANSFORMER, 190
PRETTYPRINT_UNSTRUCTURED, 191
PRETTYPRINT_UNSTRUCTURED_AS_A_GRAPH, 198
PRETTYPRINT_UNSTRUCTURED_AS_A_GRAPH_VERBOSE, 198
PRETTYPRINT_VARIABLE_DIMENSIONS, 193
PRETTYPRINT_WITH_COMMON_NAMES, 191
Prettyprinter, 176
Prettyprinter (Code), 188
Prettyprinter (DG), 50
Prettyprinter (HPF), 194
Prettyprinter Claire, 219
Prettyprinter PoCC, 218
Prettyprinters Smalltalk, 218
PRINT_DEPENDENCE_GRAPH, 50
PRINT_DEPENDENCE_GRAPH_USING_SRU_FORMAT, 50
PRINT_DEPENDENCE_GRAPH_WITH_DEPENDENCE_CONES, 50
PRINT_DEPENDENCE_GRAPH_WITHOUT_NOLOOPCARRIED_DEPS, 50
PRINT_DEPENDENCE_GRAPH_WITHOUT_PRIVATIZED_DEPS, 50
Privationzation (Array), 170
Privatization, 168, 170
Program Transformation, 121
Proper Effects, 35
Reduction, 96
Reduction Detection, 144
Reduction Parallelization, 171
Redudant Load-Store Elimination, 96
Region, 70
Region (Array), 75
Region (Summary), 73
Regions (IN), 73
Regions (OUT), 74
REGIONS_OP_STATISTICS, 75
REGIONS_TRANSLATION_STATISTICS, 75
REGIONS_WITH_ARRAY_BOUNDS, 75
RESTRUCTURE_WHILE_RECOVER, 134
Restructurer, 134
Return (Alternate), 22
RI, 18
RICE_DATAFLOW_DEPENDENCE_ONLY, 49
RICEDG_PROVIDE_STATISTICS FALSE, 48
RICEDG_STATISTICS_ALL_ARRAYS, 48
Safescale, 107
Scalar Expansion, 171
Scalar Privatization, 169
Scalar Renaming, 94
Scheduling, 221
SDFI, 37
Semantics, 60
Semantics Analysis, 51
SEMANTICS_ANALYZE_SCALAR_BOOLEAN_VARIABLES, 60
SEMANTICS_ANALYZE_SCALAR_FLOAT_VARIABLES, 60
SEMANTICS_ANALYZE_SCALAR_INTEGER_VARIABLES, 60
SEMANTICS_ANALYZE_SCALAR_STRING_VARIABLES, 60
SEMANTICS_ANALYZE_UNSTRUCTURED, 61
SEMANTICS_FILTERED_PRECONDITIONS, 65
SEMANTICS_FIX_POINT, 63
SEMANTICS_FIX_POINT_OPERATOR, 63
SEMANTICS_FLOW_SENSITIVE, 61
SEMANTICS_INEQUALITY_INVARIANT, 63
SEMANTICS_INTERPROCEDURAL, 62
SEMANTICS_NORMALIZATION_LEVEL_BEFORE_STORAGE, 65
SEMANTICS_RECOMPUTE_FIX_POINTS_WITH_PRECONDITIONS, 63
SEMANTICS_STDOUT, 65
SEMANTICS_TRUST_ARRAY_DECLARATIONS, 61
SEMANTICS_TRUST_ARRAY_REFERENCES, 61
SEMANTICS_USE_TYPE_INFORMATION, 61
Sequential View, 181
Simplify Control, 131
SLP, 94
Software Caching, 100
Source File, 15
Spaghettifier, 137
Specialize, 172
Splitting, 12
SSE, 94
Statement externalization, 108
Statement Function, 24
Statement number, 188
Statistics (Dependence test), 48
STF, 138
Strip-Mining, 127
Summary Complementary Sections, 83
Summary Complexity, 67
Summary Precondition, 56
Summary Region, 73
Summary Regions (IN), 74
Summary Regions (OUT), 74
Summary Total Postcondition, 60
Summary Total Precondition, 59
Summary Transformer, 54
Superword parallelism, 94
Symbol table, 20
Terapix, 109
Thread-safe library, 90
Three Address Code, 142
Three-Address Code, 141
Tiling, 95, 128
Top Level, 8
Total Postcondition (Summary), 60
Total Precondition, 58
Total Precondition (Summary), 59
Tpips, 10
TPIPS_IS_A_SHELL, 10
Transformation, 121
TRANSFORMATION_CLONE_ON_ARGUMENT, 158
Transformer, 51, 60, 62
Transformer (Summary), 54
Trivial Test Elimination, 139
Type Checking, 15, 21
TypeChecker, 173
Uniform Complexity, 67
Unreachable Code Elimination, 131
Unspaghettify, 134
UNSPAGHETTIFY_DISPLAY_STATISTICS, 134
UNSPAGHETTIFY_RECURSIVE_DECOMPOSITION, 134
UNSPAGHETTIFY_TEST_RESTRUCTURING, 134
Use-Def Chains, 42, 43
Use-Def Elimination, 133
Use-Use Chains, 42
User File, 11
Variable, 18
Vectorization, 91
WARN_ABOUT_EMPTY_SEQUENCES, 10
Warning, 10
WARNING_ON_STAT_ERROR, 10
WP65, 100