pax_global_header00006660000000000000000000000064145074766670014540gustar00rootroot0000000000000052 comment=ebe1a6aa68c280910ee24d386c86ab107a95e28c parany-14.0.1/000077500000000000000000000000001450747666700131155ustar00rootroot00000000000000parany-14.0.1/.gitignore000066400000000000000000000000311450747666700150770ustar00rootroot00000000000000_build .merlin *.install parany-14.0.1/.merlin000066400000000000000000000001421450747666700144010ustar00rootroot00000000000000# where are .ml and .mli files S . # where are .cmi and .cmt files B _build PKG unix netmulticore parany-14.0.1/LICENSE000066400000000000000000000632501450747666700141300ustar00rootroot00000000000000The Library is distributed under the terms of the GNU Library General Public License version 2 (included below). As a special exception to the GNU Library General Public License, you may link, statically or dynamically, a "work that uses the Library" with a publicly distributed version of the Library to produce an executable file containing portions of the Library, and distribute that executable file under terms of your choice, without any of the additional requirements listed in clause 6 of the GNU Library General Public License. By "a publicly distributed version of the Library", we mean either the unmodified Library as distributed, or a modified version of the Library that is distributed under the conditions defined in clause 3 of the GNU Library General Public License. This exception does not however invalidate any other reasons why the executable file might be covered by the GNU Library General Public License. ====================================================================== GNU LIBRARY GENERAL PUBLIC LICENSE Version 2, June 1991 Copyright (C) 1991 Free Software Foundation, Inc. 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA Everyone is permitted to copy and distribute verbatim copies of this license document, but changing it is not allowed. [This is the first released version of the library GPL. It is numbered 2 because it goes with version 2 of the ordinary GPL.] Preamble The licenses for most software are designed to take away your freedom to share and change it. By contrast, the GNU General Public Licenses are intended to guarantee your freedom to share and change free software--to make sure the software is free for all its users. This license, the Library General Public License, applies to some specially designated Free Software Foundation software, and to any other libraries whose authors decide to use it. You can use it for your libraries, too. When we speak of free software, we are referring to freedom, not price. Our General Public Licenses are designed to make sure that you have the freedom to distribute copies of free software (and charge for this service if you wish), that you receive source code or can get it if you want it, that you can change the software or use pieces of it in new free programs; and that you know you can do these things. To protect your rights, we need to make restrictions that forbid anyone to deny you these rights or to ask you to surrender the rights. These restrictions translate to certain responsibilities for you if you distribute copies of the library, or if you modify it. For example, if you distribute copies of the library, whether gratis or for a fee, you must give the recipients all the rights that we gave you. You must make sure that they, too, receive or can get the source code. If you link a program with the library, you must provide complete object files to the recipients so that they can relink them with the library, after making changes to the library and recompiling it. And you must show them these terms so they know their rights. Our method of protecting your rights has two steps: (1) copyright the library, and (2) offer you this license which gives you legal permission to copy, distribute and/or modify the library. Also, for each distributor's protection, we want to make certain that everyone understands that there is no warranty for this free library. If the library is modified by someone else and passed on, we want its recipients to know that what they have is not the original version, so that any problems introduced by others will not reflect on the original authors' reputations. Finally, any free program is threatened constantly by software patents. We wish to avoid the danger that companies distributing free software will individually obtain patent licenses, thus in effect transforming the program into proprietary software. To prevent this, we have made it clear that any patent must be licensed for everyone's free use or not licensed at all. Most GNU software, including some libraries, is covered by the ordinary GNU General Public License, which was designed for utility programs. This license, the GNU Library General Public License, applies to certain designated libraries. This license is quite different from the ordinary one; be sure to read it in full, and don't assume that anything in it is the same as in the ordinary license. The reason we have a separate public license for some libraries is that they blur the distinction we usually make between modifying or adding to a program and simply using it. Linking a program with a library, without changing the library, is in some sense simply using the library, and is analogous to running a utility program or application program. However, in a textual and legal sense, the linked executable is a combined work, a derivative of the original library, and the ordinary General Public License treats it as such. Because of this blurred distinction, using the ordinary General Public License for libraries did not effectively promote software sharing, because most developers did not use the libraries. We concluded that weaker conditions might promote sharing better. However, unrestricted linking of non-free programs would deprive the users of those programs of all benefit from the free status of the libraries themselves. This Library General Public License is intended to permit developers of non-free programs to use free libraries, while preserving your freedom as a user of such programs to change the free libraries that are incorporated in them. (We have not seen how to achieve this as regards changes in header files, but we have achieved it as regards changes in the actual functions of the Library.) The hope is that this will lead to faster development of free libraries. The precise terms and conditions for copying, distribution and modification follow. Pay close attention to the difference between a "work based on the library" and a "work that uses the library". The former contains code derived from the library, while the latter only works together with the library. Note that it is possible for a library to be covered by the ordinary General Public License rather than by this special one. GNU LIBRARY GENERAL PUBLIC LICENSE TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION 0. This License Agreement applies to any software library which contains a notice placed by the copyright holder or other authorized party saying it may be distributed under the terms of this Library General Public License (also called "this License"). Each licensee is addressed as "you". A "library" means a collection of software functions and/or data prepared so as to be conveniently linked with application programs (which use some of those functions and data) to form executables. The "Library", below, refers to any such software library or work which has been distributed under these terms. A "work based on the Library" means either the Library or any derivative work under copyright law: that is to say, a work containing the Library or a portion of it, either verbatim or with modifications and/or translated straightforwardly into another language. (Hereinafter, translation is included without limitation in the term "modification".) "Source code" for a work means the preferred form of the work for making modifications to it. For a library, complete source code means all the source code for all modules it contains, plus any associated interface definition files, plus the scripts used to control compilation and installation of the library. Activities other than copying, distribution and modification are not covered by this License; they are outside its scope. The act of running a program using the Library is not restricted, and output from such a program is covered only if its contents constitute a work based on the Library (independent of the use of the Library in a tool for writing it). Whether that is true depends on what the Library does and what the program that uses the Library does. 1. You may copy and distribute verbatim copies of the Library's complete source code as you receive it, in any medium, provided that you conspicuously and appropriately publish on each copy an appropriate copyright notice and disclaimer of warranty; keep intact all the notices that refer to this License and to the absence of any warranty; and distribute a copy of this License along with the Library. You may charge a fee for the physical act of transferring a copy, and you may at your option offer warranty protection in exchange for a fee. 2. You may modify your copy or copies of the Library or any portion of it, thus forming a work based on the Library, and copy and distribute such modifications or work under the terms of Section 1 above, provided that you also meet all of these conditions: a) The modified work must itself be a software library. b) You must cause the files modified to carry prominent notices stating that you changed the files and the date of any change. c) You must cause the whole of the work to be licensed at no charge to all third parties under the terms of this License. d) If a facility in the modified Library refers to a function or a table of data to be supplied by an application program that uses the facility, other than as an argument passed when the facility is invoked, then you must make a good faith effort to ensure that, in the event an application does not supply such function or table, the facility still operates, and performs whatever part of its purpose remains meaningful. (For example, a function in a library to compute square roots has a purpose that is entirely well-defined independent of the application. Therefore, Subsection 2d requires that any application-supplied function or table used by this function must be optional: if the application does not supply it, the square root function must still compute square roots.) These requirements apply to the modified work as a whole. If identifiable sections of that work are not derived from the Library, and can be reasonably considered independent and separate works in themselves, then this License, and its terms, do not apply to those sections when you distribute them as separate works. But when you distribute the same sections as part of a whole which is a work based on the Library, the distribution of the whole must be on the terms of this License, whose permissions for other licensees extend to the entire whole, and thus to each and every part regardless of who wrote it. Thus, it is not the intent of this section to claim rights or contest your rights to work written entirely by you; rather, the intent is to exercise the right to control the distribution of derivative or collective works based on the Library. In addition, mere aggregation of another work not based on the Library with the Library (or with a work based on the Library) on a volume of a storage or distribution medium does not bring the other work under the scope of this License. 3. You may opt to apply the terms of the ordinary GNU General Public License instead of this License to a given copy of the Library. To do this, you must alter all the notices that refer to this License, so that they refer to the ordinary GNU General Public License, version 2, instead of to this License. (If a newer version than version 2 of the ordinary GNU General Public License has appeared, then you can specify that version instead if you wish.) Do not make any other change in these notices. Once this change is made in a given copy, it is irreversible for that copy, so the ordinary GNU General Public License applies to all subsequent copies and derivative works made from that copy. This option is useful when you wish to copy part of the code of the Library into a program that is not a library. 4. You may copy and distribute the Library (or a portion or derivative of it, under Section 2) in object code or executable form under the terms of Sections 1 and 2 above provided that you accompany it with the complete corresponding machine-readable source code, which must be distributed under the terms of Sections 1 and 2 above on a medium customarily used for software interchange. If distribution of object code is made by offering access to copy from a designated place, then offering equivalent access to copy the source code from the same place satisfies the requirement to distribute the source code, even though third parties are not compelled to copy the source along with the object code. 5. A program that contains no derivative of any portion of the Library, but is designed to work with the Library by being compiled or linked with it, is called a "work that uses the Library". Such a work, in isolation, is not a derivative work of the Library, and therefore falls outside the scope of this License. However, linking a "work that uses the Library" with the Library creates an executable that is a derivative of the Library (because it contains portions of the Library), rather than a "work that uses the library". The executable is therefore covered by this License. Section 6 states terms for distribution of such executables. When a "work that uses the Library" uses material from a header file that is part of the Library, the object code for the work may be a derivative work of the Library even though the source code is not. Whether this is true is especially significant if the work can be linked without the Library, or if the work is itself a library. The threshold for this to be true is not precisely defined by law. If such an object file uses only numerical parameters, data structure layouts and accessors, and small macros and small inline functions (ten lines or less in length), then the use of the object file is unrestricted, regardless of whether it is legally a derivative work. (Executables containing this object code plus portions of the Library will still fall under Section 6.) Otherwise, if the work is a derivative of the Library, you may distribute the object code for the work under the terms of Section 6. Any executables containing that work also fall under Section 6, whether or not they are linked directly with the Library itself. 6. As an exception to the Sections above, you may also compile or link a "work that uses the Library" with the Library to produce a work containing portions of the Library, and distribute that work under terms of your choice, provided that the terms permit modification of the work for the customer's own use and reverse engineering for debugging such modifications. You must give prominent notice with each copy of the work that the Library is used in it and that the Library and its use are covered by this License. You must supply a copy of this License. If the work during execution displays copyright notices, you must include the copyright notice for the Library among them, as well as a reference directing the user to the copy of this License. Also, you must do one of these things: a) Accompany the work with the complete corresponding machine-readable source code for the Library including whatever changes were used in the work (which must be distributed under Sections 1 and 2 above); and, if the work is an executable linked with the Library, with the complete machine-readable "work that uses the Library", as object code and/or source code, so that the user can modify the Library and then relink to produce a modified executable containing the modified Library. (It is understood that the user who changes the contents of definitions files in the Library will not necessarily be able to recompile the application to use the modified definitions.) b) Accompany the work with a written offer, valid for at least three years, to give the same user the materials specified in Subsection 6a, above, for a charge no more than the cost of performing this distribution. c) If distribution of the work is made by offering access to copy from a designated place, offer equivalent access to copy the above specified materials from the same place. d) Verify that the user has already received a copy of these materials or that you have already sent this user a copy. For an executable, the required form of the "work that uses the Library" must include any data and utility programs needed for reproducing the executable from it. However, as a special exception, the source code distributed need not include anything that is normally distributed (in either source or binary form) with the major components (compiler, kernel, and so on) of the operating system on which the executable runs, unless that component itself accompanies the executable. It may happen that this requirement contradicts the license restrictions of other proprietary libraries that do not normally accompany the operating system. Such a contradiction means you cannot use both them and the Library together in an executable that you distribute. 7. You may place library facilities that are a work based on the Library side-by-side in a single library together with other library facilities not covered by this License, and distribute such a combined library, provided that the separate distribution of the work based on the Library and of the other library facilities is otherwise permitted, and provided that you do these two things: a) Accompany the combined library with a copy of the same work based on the Library, uncombined with any other library facilities. This must be distributed under the terms of the Sections above. b) Give prominent notice with the combined library of the fact that part of it is a work based on the Library, and explaining where to find the accompanying uncombined form of the same work. 8. You may not copy, modify, sublicense, link with, or distribute the Library except as expressly provided under this License. Any attempt otherwise to copy, modify, sublicense, link with, or distribute the Library is void, and will automatically terminate your rights under this License. However, parties who have received copies, or rights, from you under this License will not have their licenses terminated so long as such parties remain in full compliance. 9. You are not required to accept this License, since you have not signed it. However, nothing else grants you permission to modify or distribute the Library or its derivative works. These actions are prohibited by law if you do not accept this License. Therefore, by modifying or distributing the Library (or any work based on the Library), you indicate your acceptance of this License to do so, and all its terms and conditions for copying, distributing or modifying the Library or works based on it. 10. Each time you redistribute the Library (or any work based on the Library), the recipient automatically receives a license from the original licensor to copy, distribute, link with or modify the Library subject to these terms and conditions. You may not impose any further restrictions on the recipients' exercise of the rights granted herein. You are not responsible for enforcing compliance by third parties to this License. 11. If, as a consequence of a court judgment or allegation of patent infringement or for any other reason (not limited to patent issues), conditions are imposed on you (whether by court order, agreement or otherwise) that contradict the conditions of this License, they do not excuse you from the conditions of this License. If you cannot distribute so as to satisfy simultaneously your obligations under this License and any other pertinent obligations, then as a consequence you may not distribute the Library at all. For example, if a patent license would not permit royalty-free redistribution of the Library by all those who receive copies directly or indirectly through you, then the only way you could satisfy both it and this License would be to refrain entirely from distribution of the Library. If any portion of this section is held invalid or unenforceable under any particular circumstance, the balance of the section is intended to apply, and the section as a whole is intended to apply in other circumstances. It is not the purpose of this section to induce you to infringe any patents or other property right claims or to contest validity of any such claims; this section has the sole purpose of protecting the integrity of the free software distribution system which is implemented by public license practices. Many people have made generous contributions to the wide range of software distributed through that system in reliance on consistent application of that system; it is up to the author/donor to decide if he or she is willing to distribute software through any other system and a licensee cannot impose that choice. This section is intended to make thoroughly clear what is believed to be a consequence of the rest of this License. 12. If the distribution and/or use of the Library is restricted in certain countries either by patents or by copyrighted interfaces, the original copyright holder who places the Library under this License may add an explicit geographical distribution limitation excluding those countries, so that distribution is permitted only in or among countries not thus excluded. In such case, this License incorporates the limitation as if written in the body of this License. 13. The Free Software Foundation may publish revised and/or new versions of the Library General Public License from time to time. Such new versions will be similar in spirit to the present version, but may differ in detail to address new problems or concerns. Each version is given a distinguishing version number. If the Library specifies a version number of this License which applies to it and "any later version", you have the option of following the terms and conditions either of that version or of any later version published by the Free Software Foundation. If the Library does not specify a license version number, you may choose any version ever published by the Free Software Foundation. 14. If you wish to incorporate parts of the Library into other free programs whose distribution conditions are incompatible with these, write to the author to ask for permission. For software which is copyrighted by the Free Software Foundation, write to the Free Software Foundation; we sometimes make exceptions for this. Our decision will be guided by the two goals of preserving the free status of all derivatives of our free software and of promoting the sharing and reuse of software generally. NO WARRANTY 15. BECAUSE THE LIBRARY IS LICENSED FREE OF CHARGE, THERE IS NO WARRANTY FOR THE LIBRARY, TO THE EXTENT PERMITTED BY APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES PROVIDE THE LIBRARY "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE LIBRARY IS WITH YOU. SHOULD THE LIBRARY PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 16. IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY AND/OR REDISTRIBUTE THE LIBRARY AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR INABILITY TO USE THE LIBRARY (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD PARTIES OR A FAILURE OF THE LIBRARY TO OPERATE WITH ANY OTHER SOFTWARE), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. END OF TERMS AND CONDITIONS How to Apply These Terms to Your New Libraries If you develop a new library, and you want it to be of the greatest possible use to the public, we recommend making it free software that everyone can redistribute and change. You can do so by permitting redistribution under these terms (or, alternatively, under the terms of the ordinary General Public License). To apply these terms, attach the following notices to the library. It is safest to attach them to the start of each source file to most effectively convey the exclusion of warranty; and each file should have at least the "copyright" line and a pointer to where the full notice is found. Copyright (C) This library is free software; you can redistribute it and/or modify it under the terms of the GNU Library General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. This library is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public License for more details. You should have received a copy of the GNU Library General Public License along with this library; if not, write to the Free Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA Also add information on how to contact you by electronic and paper mail. You should also get your employer (if you work as a programmer) or your school, if any, to sign a "copyright disclaimer" for the library, if necessary. Here is a sample; alter the names: Yoyodyne, Inc., hereby disclaims all copyright interest in the library `Frob' (a library for tweaking knobs) written by James Random Hacker. , 1 April 1990 Ty Coon, President of Vice That's all there is to it! parany-14.0.1/Makefile000066400000000000000000000005311450747666700145540ustar00rootroot00000000000000.PHONY: build build-test clean edit install uninstall reinstall test build: dune build @install -j 16 build-test: build dune build src/test.exe dune build src/test_parmap.exe clean: rm -rf _build edit: emacs src/*.ml & install: build dune install uninstall: dune uninstall reinstall: uninstall install test: build-test ./test.sh parany-14.0.1/README.md000066400000000000000000000014011450747666700143700ustar00rootroot00000000000000# parany Parallelize any kind of computation Generalized map reduce for parallel computers (not distributed computing). Can process a very large file in parallel on a multicore computer; provided there is a way to cut your file into independent blocks (the "demux" function). Can process in parallel an infinite stream of elements (if you don't care about their processing order). In practice, if the stream is really infinite but you want to collect results in order, some integer counter will eventually overflow and the behavior is unspecified and unknown after this unfortunate event. The processing function is called "work". The function gathering the results is called "mux". The number of processors running your computation in parallel is called "nprocs". parany-14.0.1/bench.txt000066400000000000000000000021411450747666700147330ustar00rootroot00000000000000without sendmsg (Wed Mar 25 12:58:03 JST 2020) ============================================== time_build/default/src/test.exe 1 1 2>err 1>res # real 0m0.038s # user 0m0.037s # sys 0m0.001s time _build/default/src/test.exe 2 1 2>err 1>res # real 0m5.278s # user 0m0.748s # sys 0m3.987s time _build/default/src/test.exe 16 1 2>err 1>res # real 0m8.712s # user 0m1.056s # sys 0m7.111s ./gram -i data/tox21_nrar_ligands_std_rand_01.csv -np 8 # 2020-03-25 12:56:26.126 INFO : samples: 7026 features: 1972 # 48.00 18.00 4.00 ... 16.00 15.00 8.00 # 18.00 28.00 5.00 ... 24.00 8.00 7.00 # 4.00 5.00 40.00 ... 12.00 28.00 38.00 # ... # 16.00 24.00 12.00 ... 64.00 16.00 16.00 # 15.00 8.00 28.00 ... 16.00 54.00 34.00 # 8.00 7.00 38.00 ... 16.00 34.00 72.00 # 2020-03-25 12:57:23.527 INFO : n: 1 c: 1 s: seq dt: 57.38 a: 1.00 # 2020-03-25 12:57:39.016 INFO : n: 8 c: 1 s: parmap dt: 15.38 a: 3.73 # 2020-03-25 12:57:50.822 INFO : n: 8 c: 1 s: parany dt: 11.08 a: 5.18 parany-14.0.1/dune-project000066400000000000000000000000201450747666700154270ustar00rootroot00000000000000(lang dune 1.6) parany-14.0.1/parany.opam000066400000000000000000000030511450747666700152640ustar00rootroot00000000000000opam-version: "2.0" maintainer: "unixjunkie@sdf.org" authors: "Francois Berenger" license: "LGPL-2.0-or-later" homepage: "https://github.com/UnixJunkie/parany" bug-reports: "https://github.com/UnixJunkie/parany/issues" dev-repo: "git+https://github.com/UnixJunkie/parany.git" depends: [ "base-unix" "cpu" "dune" {>= "1.6.0"} "ocaml" {>= "4.03.0"} ] available: os != "win32" build: [ ["dune" "build" "-p" name "-j" jobs] ["dune" "build" "-p" name "-j" jobs "src/test.exe"] {with-test} ["./test.sh"] {with-test} ] synopsis: "Parallelize any computation" description: """ Generalized map reduce for parallel computers (not distributed computing). Can process in parallel an infinite stream of elements. Can process a very large file in parallel on a multicore computer; provided there is a way to cut your file into independent blocks (the 'demux' function). The processing function is called 'work'. The function gathering the results is called 'mux'. The number of processors running your computation in parallel is called 'nprocs'. The chunk size (number of items) processed by one call to the 'work' function is called 'csize'. There is a minimalist Parmap module, if you want to switch to/from Parmap easily. Read the corresponding ocamldoc before using. USING THIS LIBRARY IN A MISSION CRITICAL, LONG-RUNNING SOFTWARE THAT IS NEVER SUPPOSED TO CRASH IS NOT ADVIZED. WHILE THIS LIBRARY IS HIGH PERFORMANCE, IT IS ALSO DANGEROUS. USE AT YOUR OWN RISKS. """ #url { # src: "https://github.com/UnixJunkie/parany/archive/vXXX.tar.gz" # checksum: "md5=YYY" #} parany-14.0.1/src/000077500000000000000000000000001450747666700137045ustar00rootroot00000000000000parany-14.0.1/src/dune000066400000000000000000000002751450747666700145660ustar00rootroot00000000000000 (library (name parany) (public_name parany) (modules parany) (libraries unix cpu)) (executables (names test test_parmap) (modules test test_parmap) (libraries parany unix)) parany-14.0.1/src/parany.ml000066400000000000000000000322531450747666700155350ustar00rootroot00000000000000 open Printf module A = Array module Fn = Filename module Ht = Hashtbl exception End_of_input module Shm = struct let init () = Unix.(socketpair PF_UNIX SOCK_DGRAM 0) let unmarshal_from_file fn = let input = open_in_bin fn in let res = Marshal.from_channel input in close_in input; res let marshal_to_file fn v = let out = open_out_bin fn in Marshal.to_channel out v Marshal.[No_sharing; Closures]; close_out out let rec send_loop sock buff n = try let sent = Unix.send sock buff 0 n [] in assert(sent = n) with Unix.Unix_error(ENOBUFS, _, _) -> (* send on a UDP socket never blocks on Mac OS X and probably several of the BSDs *) (* eprintf "sleep\n%!"; *) let _ = Unix.select [] [] [] 0.001 in (* wait *) (* We should use nanosleep for precision, if only it was provided by the Unix module... *) send_loop sock buff n let raw_send sock str = let n = String.length str in let buff = Bytes.unsafe_of_string str in send_loop sock buff n let send fn queue to_send = marshal_to_file fn to_send; raw_send queue fn let raw_receive sock buff = let n = Bytes.length buff in let received = Unix.recv sock buff 0 n [] in assert(received > 0); Bytes.sub_string buff 0 received let receive queue buff = let fn = raw_receive queue buff in if fn = "EOF" then raise End_of_input else let res = unmarshal_from_file fn in Sys.remove fn; res end (* feeder process loop *) let feed_them_all tmp csize ncores demux queue = (* let pid = Unix.getpid () in * eprintf "feeder(%d) started\n%!" pid; *) let in_count = ref 0 in let prfx = Filename.temp_file ~temp_dir:tmp "iparany_" "" in let to_send = ref [] in try while true do for _ = 1 to csize do to_send := (demux ()) :: !to_send done; let fn = sprintf "%s_%d" prfx !in_count in Shm.send fn queue !to_send; (* eprintf "feeder(%d) sent one\n%!" pid; *) to_send := []; incr in_count done with End_of_input -> begin (* if needed, send remaining jobs (< csize) *) (if !to_send <> [] then let fn = sprintf "%s_%d" prfx !in_count in Shm.send fn queue !to_send); (* send an EOF to each worker *) for _ = 1 to ncores do Shm.raw_send queue "EOF" done; (* eprintf "feeder(%d) finished\n%!" pid; *) Sys.remove prfx; Unix.close queue end (* worker process loop *) let go_to_work prfx jobs_queue work results_queue = (* let pid = Unix.getpid () in * eprintf "worker(%d) started\n%!" pid; *) try let out_count = ref 0 in let buff = Bytes.create 80 in while true do let xs = Shm.receive jobs_queue buff in let ys = List.rev_map work xs in (* eprintf "worker(%d) did one\n%!" pid; *) let fn = sprintf "%s_%d" prfx !out_count in Shm.send fn results_queue ys; incr out_count done with End_of_input -> (* resource cleanup was moved to an at_exit-registered function, so that cleanup is done even in the case of an uncaught exception and the muxer doesn't enter an infinite loop *) () let fork_out f = match Unix.fork () with | -1 -> failwith "Parany.fork_out: fork failed" | 0 -> let () = f () in exit 0 | _pid -> () (* demux and index items *) let idemux (demux: unit -> 'a) = let demux_count = ref 0 in function () -> let res = (!demux_count, demux ()) in incr demux_count; res (* work ignoring item index *) let iwork (work: 'a -> 'b) ((i, x): int * 'a): int * 'b = (i, work x) (* mux items in the right order *) let imux (mux: 'b -> unit) = let mux_count = ref 0 in (* weak type variable avoidance *) let wait_list = Ht.create 11 in function (i, res) -> if !mux_count = i then begin (* unpile as much as possible *) mux res; incr mux_count; if Ht.length wait_list > 0 then try while true do let next = Ht.find wait_list !mux_count in Ht.remove wait_list !mux_count; mux next; incr mux_count done with Not_found -> () (* no more or index hole *) end else (* put somewhere into the pile *) Ht.add wait_list i res (* once initialized, my_rank will be in [0:ncores-1] *) let my_rank = ref (-1) let get_rank () = !my_rank let run ?(init = fun (_rank: int) -> ()) ?(finalize = fun () -> ()) ?(preserve = false) ?(core_pin = false) ?(csize = 1) nprocs ~demux ~work ~mux = (* the (type a b) annotation unfortunately implies OCaml >= 4.03.0 *) let demux_work_mux (type a b) ~(demux: unit -> a) ~(work: a -> b) ~(mux: b -> unit): unit = (* create queues *) let jobs_in, jobs_out = Shm.init () in let res_in, res_out = Shm.init () in (* start feeder *) (* eprintf "father(%d) starting feeder\n%!" pid; *) flush_all (); (* prevent duplicated I/O *) Gc.compact (); (* like parmap: reclaim memory prior to forking *) let tmp = Filename.get_temp_dir_name () in fork_out (fun () -> feed_them_all tmp csize nprocs demux jobs_in); (* start workers *) for worker_rank = 0 to nprocs - 1 do my_rank := worker_rank; (* eprintf "father(%d) starting a worker\n%!" pid; *) fork_out (fun () -> init worker_rank; (* per-process optional setup *) at_exit finalize; (* register optional finalize fun *) (* parmap also does core pinning _after_ having called the per-process init function *) if core_pin then Cpu.setcore worker_rank; let prfx = Filename.temp_file ~temp_dir:tmp (sprintf "oparany%d_" worker_rank) "" in at_exit (fun () -> (* tell collector to stop *) (* eprintf "worker(%d) finished\n%!" pid; *) Shm.raw_send res_in "EOF"; Unix.close res_in; Sys.remove prfx ); go_to_work prfx jobs_out work res_in ) done; (* collect results *) let finished = ref 0 in let buff = Bytes.create 80 in while !finished < nprocs do try while true do let xs = Shm.receive res_out buff in (* eprintf "father(%d) collecting one\n%!" pid; *) List.iter mux xs done with End_of_input -> incr finished done; (* free resources *) List.iter Unix.close [jobs_in; jobs_out; res_in; res_out] in if nprocs <= 1 then (* sequential version *) try while true do mux (work (demux ())) done with End_of_input -> () else begin (* parallel version *) assert(csize >= 1); let max_cores = Cpu.numcores () in assert(nprocs <= max_cores); (* let pid = Unix.getpid () in * eprintf "father(%d) started\n%!" pid; *) (if preserve then (* In some cases, it is necessary for the user to preserve the input order in the output. In this case, we still compute things potentially out of order (for parallelization efficiency); but we will order back the results in input order (for user's convenience) *) demux_work_mux ~demux:(idemux demux) ~work:(iwork work) ~mux:(imux mux) else (* by default, to maximize parallel efficiency we don't care about the order in which jobs are computed. *) demux_work_mux ~demux ~work ~mux ); (* eprintf "father(%d) finished\n%!" pid; *) end (* Wrapper for near-compatibility with Parmap *) module Parmap = struct let tail_rec_map f l = List.rev (List.rev_map f l) let tail_rec_mapi f l = let i = ref 0 in let res = List.rev_map (fun x -> let j = !i in let y = f j x in incr i; y ) l in List.rev res let parmap ?(init = fun (_rank: int) -> ()) ?(finalize = fun () -> ()) ?(preserve = false) ?(core_pin = false) ?(csize = 1) ncores f l = if ncores <= 1 then tail_rec_map f l else let input = ref l in let demux () = match !input with | [] -> raise End_of_input | x :: xs -> (input := xs; x) in let output = ref [] in let mux x = output := x :: !output in (* parallel work *) run ~init ~finalize ~preserve ~core_pin ~csize ncores ~demux ~work:f ~mux; if preserve then List.rev !output else !output let parmapi ?(init = fun (_rank: int) -> ()) ?(finalize = fun () -> ()) ?(preserve = false) ?(core_pin = false) ?(csize = 1) ncores f l = if ncores <= 1 then tail_rec_mapi f l else let input = ref l in let i = ref 0 in let demux () = match !input with | [] -> raise End_of_input | x :: xs -> begin let j = !i in input := xs; let res = (j, x) in incr i; res end in let output = ref [] in let f' (i, x) = f i x in let mux x = output := x :: !output in (* parallel work *) run ~init ~finalize ~preserve ~core_pin ~csize ncores ~demux ~work:f' ~mux; if preserve then List.rev !output else !output let pariter ?(init = fun (_rank: int) -> ()) ?(finalize = fun () -> ()) ?(preserve = false) ?(core_pin = false) ?(csize = 1) ncores f l = if ncores <= 1 then List.iter f l else let input = ref l in let demux () = match !input with | [] -> raise End_of_input | x :: xs -> (input := xs; x) in (* parallel work *) run ~init ~finalize ~preserve ~core_pin ~csize ncores ~demux ~work:f ~mux:ignore let parfold ?(init = fun (_rank: int) -> ()) ?(finalize = fun () -> ()) ?(preserve = false) ?(core_pin = false) ?(csize = 1) ncores f g init_acc l = if ncores <= 1 then List.fold_left (fun acc x -> g acc (f x)) init_acc l else let input = ref l in let demux () = match !input with | [] -> raise End_of_input | x :: xs -> (input := xs; x) in let output = ref init_acc in let mux x = output := g !output x in (* parallel work *) run ~init ~finalize ~preserve ~core_pin ~csize ncores ~demux ~work:f ~mux; !output (* preserves array input order *) let array_parmap ?(init = fun (_rank: int) -> ()) ?(finalize = fun () -> ()) ?(core_pin = false) ?(csize = 1) ncores f init_acc a = let n = A.length a in let res = A.make n init_acc in run ~init ~finalize ~preserve:false (* input-order is preserved explicitely below *) ~core_pin ~csize ncores ~demux:( let in_count = ref 0 in fun () -> if !in_count = n then raise End_of_input else let i = !in_count in incr in_count; i) ~work:(fun i -> (i, f (A.unsafe_get a i))) ~mux:(fun (i, y) -> A.unsafe_set res i y); res let array_pariter ?(init = fun (_rank: int) -> ()) ?(finalize = fun () -> ()) ?(core_pin = false) ?(csize = 1) ncores f a = run ~init ~finalize ~preserve:false ~core_pin ~csize ncores ~demux:( let n = A.length a in let in_count = ref 0 in fun () -> if !in_count = n then raise End_of_input else let i = !in_count in incr in_count; i) ~work:(fun i -> f (A.unsafe_get a i)) ~mux:(fun () -> ()) let array_pariteri ?(init = fun (_rank: int) -> ()) ?(finalize = fun () -> ()) ?(core_pin = false) ?(csize = 1) ncores f a = run ~init ~finalize ~preserve:false ~core_pin ~csize ncores ~demux:( let n = A.length a in let in_count = ref 0 in fun () -> if !in_count = n then raise End_of_input else let i = !in_count in incr in_count; i) ~work:(fun i -> f i (A.unsafe_get a i)) ~mux:(fun () -> ()) (* let parfold_compat * ?(init = fun (_rank: int) -> ()) ?(finalize = fun () -> ()) * ?(ncores: int option) ?(chunksize: int option) (f: 'a -> 'b -> 'b) * (l: 'a list) (init_acc: 'b) (acc_fun: 'b -> 'b -> 'b): 'b = * let nprocs = match ncores with * | None -> 1 (\* if the user doesn't know the number of cores to use, * we don't know better *\) * | Some x -> x in * let csize = match chunksize with * | None -> 1 * | Some x -> x in * if nprocs <= 1 then * List.fold_left (fun acc x -> f x acc) init_acc l * else * let input = ref l in * let demux () = match !input with * | [] -> raise End_of_input * | _ -> * let this_chunk, rest = BatList.takedrop csize !input in * input := rest; * this_chunk in * let work xs = * List.fold_left (fun acc x -> f x acc) init_acc xs in * let output = ref init_acc in * let mux x = * output := acc_fun !output x in * (\* parallel work *\) * run ~init ~finalize * (\* leave csize=1 bellow *\) * ~preserve:false ~core_pin:false ~csize:1 nprocs ~demux ~work ~mux; * !output *) end parany-14.0.1/src/parany.mli000066400000000000000000000072761450747666700157150ustar00rootroot00000000000000 (** The [demux] function must throw [Parany.End_of_input] once it is done. *) exception End_of_input (** [run ~csize:10 16 ~demux:f ~work:g ~mux:h] will run in parallel on 16 cores the [g] function. Inputs to function [g] are produced by function [f] and grouped by 10 (the chunk size [csize]). If not provided, [csize] defaults to one. The performance-optimal [csize] depends on your computer, the functions you are using and the granularity of your computation. Elements which are fast to process may benefit from a [csize] greater than one. The demux function [f] must throw [Parany.End_of_input] once it is done. Outputs of function [g] are consumed by function [h]. Functions [f] and [g] are run by different unix processes. Function [g] is run by several processes at the same time (16 in this example). Only function [mux] is run by the same process that called [Parany.run]. [~core_pin] is an optional parameter which defaults to false. Core pinning can improve performance but should not be used on computers with many users or running several parany computations at the same time. [~preserve] is an optional parameter which defaults to false. If set to true, results will be accumulated by [h] in the same order that function [f] emitted them. However, for parallel performance reasons, the jobs are still potentially computed by [g] out of order. The optional [init] (resp. [finalize]) function is called once by each child process just after creation (resp. just before exit). [init] and [finalize] both default to doing nothing. [init i] takes the child rank [i] as parameter (first forked child process has rank 0, next 1, etc.). *) val run: ?init:(int -> unit) -> ?finalize:(unit -> unit) -> ?preserve:bool -> ?core_pin:bool -> ?csize:int -> int -> demux:(unit -> 'a) -> work:('a -> 'b) -> mux:('b -> unit) -> unit (** A worker process (inside the ~work function) can call [get_rank ()] to know its rank. Ranks are in [0:ncores-1] *) val get_rank: unit -> int (** Wrapper module for near-compatibility with Parmap *) module Parmap: sig (** Parallel List.map *) val parmap: ?init:(int -> unit) -> ?finalize:(unit -> unit) -> ?preserve:bool -> ?core_pin:bool -> ?csize:int -> int -> ('a -> 'b) -> 'a list -> 'b list (** Parallel List.mapi *) val parmapi: ?init:(int -> unit) -> ?finalize:(unit -> unit) -> ?preserve:bool -> ?core_pin:bool -> ?csize:int -> int -> (int -> 'a -> 'b) -> 'a list -> 'b list (** Parallel List.iter *) val pariter: ?init:(int -> unit) -> ?finalize:(unit -> unit) -> ?preserve:bool -> ?core_pin:bool -> ?csize:int -> int -> ('a -> unit) -> 'a list -> unit (** Parallel List.fold *) val parfold: ?init:(int -> unit) -> ?finalize:(unit -> unit) -> ?preserve:bool -> ?core_pin:bool -> ?csize:int -> int -> ('a -> 'b) -> ('acc -> 'b -> 'acc) -> 'acc -> 'a list -> 'acc (** Parallel Array.map; array input order is always preserved. *) val array_parmap: ?init:(int -> unit) -> ?finalize:(unit -> unit) -> ?core_pin:bool -> ?csize:int -> int -> ('a -> 'b) -> 'b -> 'a array -> 'b array (** Parallel Array.iter *) val array_pariter: ?init:(int -> unit) -> ?finalize:(unit -> unit) -> ?core_pin:bool -> ?csize:int -> int -> ('a -> unit) -> 'a array -> unit (** Parallel Array.iteri *) val array_pariteri: ?init:(int -> unit) -> ?finalize:(unit -> unit) -> ?core_pin:bool -> ?csize:int -> int -> (int -> 'a -> unit) -> 'a array -> unit (* val parfold_compat: ?init:(int -> unit) -> ?finalize:(unit -> unit) -> * ?ncores:int -> ?chunksize:int -> ('a -> 'b -> 'b) -> 'a list -> * 'b -> ('b -> 'b -> 'b) -> 'b *) end parany-14.0.1/src/test.ml000066400000000000000000000027331450747666700152220ustar00rootroot00000000000000 open Printf let n = 100_000 let inputs = Array.init n (fun i -> i) let counter = ref 0 let demux () = if !counter = n then raise Parany.End_of_input else let res = inputs.(!counter) in incr counter; res let res_counter = ref 0 let results = Array.make n 0 let mux x = results.(!res_counter) <- x; incr res_counter let bool_of_string = function | "0" -> false | "1" -> true | x -> failwith (sprintf "Test.bool_of_string: %s neither 0 nor 1" x) let rank = ref (-1) let first_time = ref true let hello i = rank := i; eprintf "I am %d\n%!" i let goodbye () = eprintf "Goodbye from %d\n%!" !rank let work a = (* if !first_time then * begin * eprintf "My rank is %d\n%!" (Parany.get_rank ()); * first_time := false * end; *) (* if !rank = 8 then failwith "Worker 8 on strike" else *) a let main () = let argc = Array.length Sys.argv in if argc <> 4 then (eprintf "usage: %s nprocs csize {0|1}\n" Sys.argv.(0); exit 1); let nprocs = int_of_string Sys.argv.(1) in let csize = int_of_string Sys.argv.(2) in let preserve = bool_of_string Sys.argv.(3) in (* Parany.set_core_pinning true; *) (if preserve then (* we test init/finalize only with preserve *) Parany.run ~init:hello ~finalize:goodbye ~preserve:true ~csize:csize nprocs ~demux ~work ~mux else Parany.run ~csize:csize nprocs ~demux ~work ~mux); for i = 0 to n - 1 do Printf.printf "%d\n" results.(i) done let () = main () parany-14.0.1/src/test_parmap.ml000066400000000000000000000011621450747666700165550ustar00rootroot00000000000000 (* test preserve parameter in Parany.Parmap.parmap{i} *) let main () = let input = [1;2;3;4;5;6;7;8] in let id x = x in let idi _i x = x in let ncores = 3 in let output1 = Parany.Parmap.parmap ~preserve:true ncores id input in assert(input = output1); let output2 = Parany.Parmap.parmap ~preserve:false ncores id input in assert(input = List.sort compare output2); let output3 = Parany.Parmap.parmapi ~preserve:true ncores idi input in assert(input = output3); let output4 = Parany.Parmap.parmapi ~preserve:false ncores idi input in assert(input = List.sort compare output4) let () = main () parany-14.0.1/test.sh000077500000000000000000000011521450747666700144320ustar00rootroot00000000000000#!/bin/bash #set -x # DEBUG NPROCS=`getconf _NPROCESSORS_ONLN` _build/default/src/test_parmap.exe # parallel with csize > 1 diff <(seq 0 99999) \ <(_build/default/src/test.exe $NPROCS 800 0 2>/dev/null | sort -n) # parallel with csize = 1 diff <(seq 0 99999) \ <(_build/default/src/test.exe $NPROCS 1 0 2>/dev/null | sort -n) # parallel with csize = 1 and preserve # DO NOT IGNORE STDERR because testing init/finalize with preserve diff <(seq 0 99999) \ <(_build/default/src/test.exe $NPROCS 1 1) # sequential diff <(seq 0 99999) \ <(_build/default/src/test.exe 1 1 0 2>/dev/null | sort -n)