- Pharo-dev - lists.pharo.org

[Pharo 8.0] Build #234: 3218-Clean-Refactoring-change-tests

by ci-pharo-ci-jenkins2＠inria.fr

There is a new Pharo build available! The status of the build #234 was: SUCCESS. The Pull Request #3221 was integrated: "3218-Clean-Refactoring-change-tests" Pull request url: https://github.com/pharo-project/pharo/pull/3221 Issue Url: https://github.com/pharo-project/pharo/issues/3218 Build Url: https://ci.inria.fr/pharo-ci-jenkins2/job/Test%20pending%20pull%20request%2…

April 25, 2019

[Pharo 8.0] Build #233: 3224-Fix-ProperMethodCategorizationTest-to-not-enforce-running-protocol-

by ci-pharo-ci-jenkins2＠inria.fr

There is a new Pharo build available! The status of the build #233 was: FAILURE. The Pull Request #3227 was integrated: "3224-Fix-ProperMethodCategorizationTest-to-not-enforce-running-protocol-" Pull request url: https://github.com/pharo-project/pharo/pull/3227 Issue Url: https://github.com/pharo-project/pharo/issues/3224 Build Url: https://ci.inria.fr/pharo-ci-jenkins2/job/Test%20pending%20pull%20request%2…

April 25, 2019

Re: [Pharo-dev] [vwnc] most energy efficient AI system - challenge of the German Federal Ministry of Education and Research BMBF

by Bernhard HÃ¶fner

Hello group, Bernard Portier from University Brest allowed me to forward his ideas referring to works of different scientists. It may help to concretize an AI-project or to fertilize discussion. I am not an AI-specialist and sorry not to be able to participate on a discussion. Best Regards, Bernhard â¦ Dear Bernhard, This is a summary of work done at University of Brest for mapping Smalltalk computation to hardware. 1) efficiency is concurrency in execution During decades researchers and industry was running behind the work of Xerox PARC on micro- programmed computers, such as the Dorado. The Dorado implemented a pipeline for decoding languages intermediate code developped for this purpose. Smalltalk bytecodes as delivered for Smalltalk-80 were of this kind. Behind the streaming engine that prepared the execution, there was the micro-machine routines translated by PARC into micro-program control executing several internal operations in parallel. The blue book implementation part is just a description of how the instructions are executed. (For those who do not know about micro-programming, there are primers written in the 70's by Michael Flynn to explain link between the instruction set, micro-program structure, and program execution. It takes 1 day to read, understand, and experiment: look at Ci lines that activate concurrent operations from a micro-code word.) [Flynn] Micro-programming, includes Rosin machine, in Computer Architecture 1972. Annex 1. [Dorado] Retrospective, Pier. http://bitsavers.trailing-edge.com/pdf/xerox/parc/techReports/ISL-83 <http://bitsavers.trailing-edge.com/pdf/xerox/parc/techReports/ISL-83>- 1_A_Retrospective_on_the_Dorado_A_High-Performance_Personal_Computer.pdf So, Smalltalk was fast enough in 1980, it was well understood for a machine execution, and became so slow when executed on micro processors, despite valuable efforts at UCBerkeley, Stanford and others. 2) concurrency is now easily available Probably one could implement Rosin machine in few days, plus a compiler to translate a calculator syntax into instruction set for it. Research could rather examine chances to execute high level computer language at the current level of resources as FPGAs devices and shared memory accelerators. In the second case, one can recall that previous generation of massive parallel computers used Smalltalk to model execution. It was the case for Maspar machine which debugger was showing clearly a ParcPlace support to display assertions on parallel (plural) variables. This debugger inspired some work to support parallel execution on GPU as shown in [VanLong]. [VanLong] Master report. Annex 2. http://wsn.univ-brest.fr/pottier/long.pdf <http://wsn.univ-brest.fr/pottier/long.pdf> (See also https://www.mdpi.com/1424-8220/18/7/2323 <https://www.mdpi.com/1424-8220/18/7/2323> about parallel physical simulation) 3) execution and definition of object spaces can replace types. Smalltalk does not have types. Types ease program checking and preparation of execution on machines. Instead, Smalltalk provides classes to resolve call to methods. Instead of relying on class tags, we can also rely on object sets (called contexts), solve binding on these object sets, and produce logic networks that will replace the slow interpretation or execution by one call to a dedicated function block. Computer operative parts are either logic networks, combinational or sequential, defined at boolean level. Thus this schema is the common way to produce operative blocks, even dynamically. The first stage is a high level look-up table tree production organized around method calls, with evolved backtracking memory optimizers. This could be suitable for architectures such as GPUs or memory based nano-architectures. The second stage is mapping to reconfigurable architectures, expecially those oriented to logic, such as FPGAs. The Madeo framework address Smalltalk to technology translation in its different stages address this question, such as: [FCCM96] Smalltalk blocks to logic, 4th IEEE Symposium on FPGAs for Custom Computing Machines (FCCM '96), Napa Valley, CA, USA, April 17-19, 1996. IEEE 1996, ISBN 0-8186-7548-9 also at http://wsn.univ-brest.fr/reports/96/st80-2-FPGA.ps.Z <http://wsn.univ-brest.fr/reports/96/st80-2-FPGA.ps.Z> [SAMOS2002] A LUT based approach for high level synthesis, Shuvra S. Bhattacharyya, E.F. Deprettere, Jurgen Teich, (Eds.), Domain-Specific Processors: Systems, Architectures, Modeling, and Simulation, Hardcover, pp. 261, plus XV, Marcel Dekker, Inc., New York, 2004, ISBN 0-8427-4711-9 also at http://wsn.univ-brest.fr/reports/2002/samos2.pdf <http://wsn.univ-brest.fr/reports/2002/samos2.pdf> 4) Mapping LUT systems to hardware were addressed by two excellent PhD works, one that draws a flattened LUT tree on an FPGA, the other one that address nano technologies, largely connected to Pr Moritz work at Umass. As a conclusion (I hope it is not final), there are large opportunities to improve specific execution speed of Smalltalk engines using new 'back to the future' approaches. We just need to stop doing bad copies of the sacred 40 years implementation, pretending that it is a 'virtual machine' for today, look to GPUs, reconfigurable devices, and new technologies especially those based on low energy and fast physical memories. Bernard Pottier UniversiteÌ de Brest, France pottier(a)univ-brest.fr <mailto:pottier@univ-brest.fr> > Am 17.04.2019 um 01:00 schrieb Sean Glazier <sglazier456(a)gmail.com <mailto:sglazier456@gmail.com>>: > > This sounds great do they need a great Smalltalker? My linked in profile is www.linkedin.com/in/seanglazier <http://www.linkedin.com/in/seanglazier> I would love to be a contributor > > Kind Regards, > > Sean Glazier > > On Mon, Apr 15, 2019 at 4:33 PM Bernhard HÃ¶fner <bernhard.hoefner(a)web.de <mailto:bernhard.hoefner@web.de>> wrote: > Thank you, Helge, I had no clue but I felt something like this. > â¦at least it would be an excellent advertisement for Smalltalk when it succeeds and therefore maybe it would be worth to harness the forces in the Smalltalk community, no matter if Squeak, Pharo, VisualWorksâ¦ > But Smalltalk will not have any effect without an excellent partner for hardware and for AI algorithms. Maybe the suitable Fraunhofer institute? At least they invented mp3. > > Best regards, Bernhard > > >> Am 15.04.2019 um 17:54 schrieb Nowak, Helge <HNowak(a)cincom.com <mailto:HNowak@cincom.com>>: >> >> Dear Bernhard, Alexandre and all, <> >> >> Thanks Bernhard for pointing to this. I would have missed it completelyâ¦ >> >> The German BMBF (Bundesministerium fÃ¼r Bildung und Forschung = German Federal Ministry of Research and Technology) started this competition for German universities and public research centers. It is targeted at increasing energy efficiency for AI systems. The aim is to increase Germanyâs competencies and competitiveness in this sector which is deemed to be a future key technology with the potential of disruptive innovation. >> >> It is clear that the energy efficiency is less determined by what programming language is used and what virtual machine is run on some standard hardware. The real gains are rather found in maybe novel algorithms, surely in novel hardware architectures and implementations. Indeed, when you look at the details of the competition (link below) that is exactly the case: any proposed solutions submitted to the competition must belong to one of three categories: FPGAs, ASIC s in 130 nm technology or ASICs in FDSOI technology. Close hardware/software co-design is explicitly requested. Not eligible are solutions that only optimize software components or use commercially available hardware unchanged. >> >> Does this mean that this competition is of no interest to ambitious Smalltalkers? No! In contrary. With Smalltalk being a perfect interactive modeling, exploration and simulation tool we Smalltalkers can help our competition team to build, explore and optimize their ideas before and while building them in hardware. Also hardware is nothing without software. Smalltalk is in my view best positioned to execute the desired co-design in a co-evolutionary way. Certainly we could demonstrate modern âagileâ engineering approaches as a side-means to increase competitiveness. >> >> Everyone interested should take a look at the competition here:https://www.bmbf.de/foerderungen/bekanntmachung-2371.html <https://www.bmbf.de/foerderungen/bekanntmachung-2371.html> (in German, obviously). >> >> Please also forward to all people who might be interested. >> >> Cheers >> Helge >> >> Helge Nowak >> Cincom Smalltalk Technical Account Manager >> <image001.png> <http://www.cincomsmalltalk.com/> >> >> Cincom Systems GmbH & Co. oHG >> HumboldtstraÃe 3 >> 60318 Frankfurt am Main >> GERMANY > >> office >> mobile >> >> website >> email >> +49 89 89 66 44 94 >> +49 172 74 00 402 >> >> http://www.cincomsmalltalk.com <http://www.cincomsmalltalk.com/> >> hnowak(a)cincom.com <mailto:hnowak@cincom.com> >> A standpoint is an intellectual horizon of radius zero. -- Albert Einstein >> >> GeschÃ¤ftsfÃ¼hrer/Managing Directors: Thomas M. Nies, Donald E. Vick >> oHG mit Sitz/based in Frankfurt am Main (Amtsgericht Frankfurt am Main HRA 50297) >> Pers. haftender Gesellschafter/Partner liable to unlimited extent: >> Cincom Systems Verwaltungsgesellschaft mbH (Amtsgericht KÃ¶nigstein/Ts. HRB 5069) >> >> --- CONFIDENTIALITY STATEMENT --- >> This e-mail transmission contains information that is intended to be privileged and confidential. It is intended only for the addressee named above. If you receive this e-mail in error, please do not read, copy or disseminate it in any manner. If you are not the intended recipient, any disclosure, copying, distribution or use of the contents of this information is prohibited, please reply to the message immediately by informing the sender that the message was misdirected. After replying, please erase it from your computer system. Your assistance in correcting this error is appreciated. > >> >> >> Von: Alexandre Bergel [mailto:alexandre.bergel@me.com <mailto:alexandre.bergel@me.com>] >> Gesendet: Freitag, 12. April 2019 19:08 >> An: Bernhard HÃ¶fner <bernhard.hoefner(a)web.de <mailto:bernhard.hoefner@web.de>> >> Cc: VisualWorks mailing list list <vwnc(a)cs.uiuc.edu <mailto:vwnc@cs.uiuc.edu>>; pharo-dev(a)lists.pharo.org <mailto:pharo-dev@lists.pharo.org> >> Betreff: Re: [vwnc] most energy efficient AI system - challenge of the German Federal Ministry of Education and Research BMBF >> >> Hello, >> >> We conducted some experiment in characterizing virtual machine from an energetic point of view. Results are not public yet. This is just to say this is a topic that I am interested in. >> >> Cheers, >> Alexandre >> >> >> On Apr 11, 2019, at 5:29 PM, Bernhard HÃ¶fner <bernhard.hoefner(a)web.de <mailto:bernhard.hoefner@web.de>> wrote: >> >> Hello Group, >> the German BMBF advertised a reward for the most energy efficient AI system, see: https://www.bmbf.de/de/mehr-ki-erfordert-weniger-energie-8156.html <https://www.bmbf.de/de/mehr-ki-erfordert-weniger-energie-8156.html> (in German), and I am wondering what is the position of Smalltalk in that field or what it could be. From the early 90th I know that much of the pretended slowness of Smalltalk was absorbed by the more efficient programs due to better analysis, concepts, design and development tools (and maybe developers ? ). >> >> Best Regards, Bernhard > > -- > > Kind Regards, > > Sean Glazier > 603 892 0167 cell > 603 583 4575 Skype phone > Skype Id: visualwave >

April 24, 2019

Re: [Pharo-dev] [vwnc] most energy efficient AI system - challenge of the German Federal Ministry of Education and Research BMBF

by Bernhard HÃ¶fner

Hello group, Bernard Portier from University Brest allowed me to forward his ideas referring to works of different scientists. It may help to concretize an AI-project or to fertilize discussion. I am not an AI-specialist and sorry not to be able to participate on a discussion. Best Regards, Bernhard â¦ Dear Bernhard, This is a summary of work done at University of Brest for mapping Smalltalk computation to hardware. 1) efficiency is concurrency in execution During decades researchers and industry was running behind the work of Xerox PARC on micro- programmed computers, such as the Dorado. The Dorado implemented a pipeline for decoding languages intermediate code developped for this purpose. Smalltalk bytecodes as delivered for Smalltalk-80 were of this kind. Behind the streaming engine that prepared the execution, there was the micro-machine routines translated by PARC into micro-program control executing several internal operations in parallel. The blue book implementation part is just a description of how the instructions are executed. (For those who do not know about micro-programming, there are primers written in the 70's by Michael Flynn to explain link between the instruction set, micro-program structure, and program execution. It takes 1 day to read, understand, and experiment: look at Ci lines that activate concurrent operations from a micro-code word.) [Flynn] Micro-programming, includes Rosin machine, in Computer Architecture 1972. Annex 1. [Dorado] Retrospective, Pier. http://bitsavers.trailing-edge.com/pdf/xerox/parc/techReports/ISL-83 <http://bitsavers.trailing-edge.com/pdf/xerox/parc/techReports/ISL-83>- 1_A_Retrospective_on_the_Dorado_A_High-Performance_Personal_Computer.pdf So, Smalltalk was fast enough in 1980, it was well understood for a machine execution, and became so slow when executed on micro processors, despite valuable efforts at UCBerkeley, Stanford and others. 2) concurrency is now easily available Probably one could implement Rosin machine in few days, plus a compiler to translate a calculator syntax into instruction set for it. Research could rather examine chances to execute high level computer language at the current level of resources as FPGAs devices and shared memory accelerators. In the second case, one can recall that previous generation of massive parallel computers used Smalltalk to model execution. It was the case for Maspar machine which debugger was showing clearly a ParcPlace support to display assertions on parallel (plural) variables. This debugger inspired some work to support parallel execution on GPU as shown in [VanLong]. [VanLong] Master report. Annex 2. http://wsn.univ-brest.fr/pottier/long.pdf <http://wsn.univ-brest.fr/pottier/long.pdf> (See also https://www.mdpi.com/1424-8220/18/7/2323 <https://www.mdpi.com/1424-8220/18/7/2323> about parallel physical simulation) 3) execution and definition of object spaces can replace types. Smalltalk does not have types. Types ease program checking and preparation of execution on machines. Instead, Smalltalk provides classes to resolve call to methods. Instead of relying on class tags, we can also rely on object sets (called contexts), solve binding on these object sets, and produce logic networks that will replace the slow interpretation or execution by one call to a dedicated function block. Computer operative parts are either logic networks, combinational or sequential, defined at boolean level. Thus this schema is the common way to produce operative blocks, even dynamically. The first stage is a high level look-up table tree production organized around method calls, with evolved backtracking memory optimizers. This could be suitable for architectures such as GPUs or memory based nano-architectures. The second stage is mapping to reconfigurable architectures, expecially those oriented to logic, such as FPGAs. The Madeo framework address Smalltalk to technology translation in its different stages address this question, such as: [FCCM96] Smalltalk blocks to logic, 4th IEEE Symposium on FPGAs for Custom Computing Machines (FCCM '96), Napa Valley, CA, USA, April 17-19, 1996. IEEE 1996, ISBN 0-8186-7548-9 also at http://wsn.univ-brest.fr/reports/96/st80-2-FPGA.ps.Z <http://wsn.univ-brest.fr/reports/96/st80-2-FPGA.ps.Z> [SAMOS2002] A LUT based approach for high level synthesis, Shuvra S. Bhattacharyya, E.F. Deprettere, Jurgen Teich, (Eds.), Domain-Specific Processors: Systems, Architectures, Modeling, and Simulation, Hardcover, pp. 261, plus XV, Marcel Dekker, Inc., New York, 2004, ISBN 0-8427-4711-9 also at http://wsn.univ-brest.fr/reports/2002/samos2.pdf <http://wsn.univ-brest.fr/reports/2002/samos2.pdf> 4) Mapping LUT systems to hardware were addressed by two excellent PhD works, one that draws a flattened LUT tree on an FPGA, the other one that address nano technologies, largely connected to Pr Moritz work at Umass. As a conclusion (I hope it is not final), there are large opportunities to improve specific execution speed of Smalltalk engines using new 'back to the future' approaches. We just need to stop doing bad copies of the sacred 40 years implementation, pretending that it is a 'virtual machine' for today, look to GPUs, reconfigurable devices, and new technologies especially those based on low energy and fast physical memories. Bernard Pottier UniversiteÌ de Brest, France pottier(a)univ-brest.fr <mailto:pottier@univ-brest.fr> the text as pdf: > Am 17.04.2019 um 01:00 schrieb Sean Glazier <sglazier456(a)gmail.com>: > > This sounds great do they need a great Smalltalker? My linked in profile is www.linkedin.com/in/seanglazier <http://www.linkedin.com/in/seanglazier> I would love to be a contributor > > Kind Regards, > > Sean Glazier > > On Mon, Apr 15, 2019 at 4:33 PM Bernhard HÃ¶fner <bernhard.hoefner(a)web.de <mailto:bernhard.hoefner@web.de>> wrote: > Thank you, Helge, I had no clue but I felt something like this. > â¦at least it would be an excellent advertisement for Smalltalk when it succeeds and therefore maybe it would be worth to harness the forces in the Smalltalk community, no matter if Squeak, Pharo, VisualWorksâ¦ > But Smalltalk will not have any effect without an excellent partner for hardware and for AI algorithms. Maybe the suitable Fraunhofer institute? At least they invented mp3. > > Best regards, Bernhard > > >> Am 15.04.2019 um 17:54 schrieb Nowak, Helge <HNowak(a)cincom.com <mailto:HNowak@cincom.com>>: >> >> Dear Bernhard, Alexandre and all, <> >> >> Thanks Bernhard for pointing to this. I would have missed it completelyâ¦ >> >> The German BMBF (Bundesministerium fÃ¼r Bildung und Forschung = German Federal Ministry of Research and Technology) started this competition for German universities and public research centers. It is targeted at increasing energy efficiency for AI systems. The aim is to increase Germanyâs competencies and competitiveness in this sector which is deemed to be a future key technology with the potential of disruptive innovation. >> >> It is clear that the energy efficiency is less determined by what programming language is used and what virtual machine is run on some standard hardware. The real gains are rather found in maybe novel algorithms, surely in novel hardware architectures and implementations. Indeed, when you look at the details of the competition (link below) that is exactly the case: any proposed solutions submitted to the competition must belong to one of three categories: FPGAs, ASIC s in 130 nm technology or ASICs in FDSOI technology. Close hardware/software co-design is explicitly requested. Not eligible are solutions that only optimize software components or use commercially available hardware unchanged. >> >> Does this mean that this competition is of no interest to ambitious Smalltalkers? No! In contrary. With Smalltalk being a perfect interactive modeling, exploration and simulation tool we Smalltalkers can help our competition team to build, explore and optimize their ideas before and while building them in hardware. Also hardware is nothing without software. Smalltalk is in my view best positioned to execute the desired co-design in a co-evolutionary way. Certainly we could demonstrate modern âagileâ engineering approaches as a side-means to increase competitiveness. >> >> Everyone interested should take a look at the competition here:https://www.bmbf.de/foerderungen/bekanntmachung-2371.html <https://www.bmbf.de/foerderungen/bekanntmachung-2371.html> (in German, obviously). >> >> Please also forward to all people who might be interested. >> >> Cheers >> Helge >> >> Helge Nowak >> Cincom Smalltalk Technical Account Manager >> <image001.png> <http://www.cincomsmalltalk.com/> >> >> Cincom Systems GmbH & Co. oHG >> HumboldtstraÃe 3 >> 60318 Frankfurt am Main >> GERMANY > >> office >> mobile >> >> website >> email >> +49 89 89 66 44 94 >> +49 172 74 00 402 >> >> http://www.cincomsmalltalk.com <http://www.cincomsmalltalk.com/> >> hnowak(a)cincom.com <mailto:hnowak@cincom.com> >> A standpoint is an intellectual horizon of radius zero. -- Albert Einstein >> >> GeschÃ¤ftsfÃ¼hrer/Managing Directors: Thomas M. Nies, Donald E. Vick >> oHG mit Sitz/based in Frankfurt am Main (Amtsgericht Frankfurt am Main HRA 50297) >> Pers. haftender Gesellschafter/Partner liable to unlimited extent: >> Cincom Systems Verwaltungsgesellschaft mbH (Amtsgericht KÃ¶nigstein/Ts. HRB 5069) >> >> --- CONFIDENTIALITY STATEMENT --- >> This e-mail transmission contains information that is intended to be privileged and confidential. It is intended only for the addressee named above. If you receive this e-mail in error, please do not read, copy or disseminate it in any manner. If you are not the intended recipient, any disclosure, copying, distribution or use of the contents of this information is prohibited, please reply to the message immediately by informing the sender that the message was misdirected. After replying, please erase it from your computer system. Your assistance in correcting this error is appreciated. > >> >> >> Von: Alexandre Bergel [mailto:alexandre.bergel@me.com <mailto:alexandre.bergel@me.com>] >> Gesendet: Freitag, 12. April 2019 19:08 >> An: Bernhard HÃ¶fner <bernhard.hoefner(a)web.de <mailto:bernhard.hoefner@web.de>> >> Cc: VisualWorks mailing list list <vwnc(a)cs.uiuc.edu <mailto:vwnc@cs.uiuc.edu>>; pharo-dev(a)lists.pharo.org <mailto:pharo-dev@lists.pharo.org> >> Betreff: Re: [vwnc] most energy efficient AI system - challenge of the German Federal Ministry of Education and Research BMBF >> >> Hello, >> >> We conducted some experiment in characterizing virtual machine from an energetic point of view. Results are not public yet. This is just to say this is a topic that I am interested in. >> >> Cheers, >> Alexandre >> >> >> On Apr 11, 2019, at 5:29 PM, Bernhard HÃ¶fner <bernhard.hoefner(a)web.de <mailto:bernhard.hoefner@web.de>> wrote: >> >> Hello Group, >> the German BMBF advertised a reward for the most energy efficient AI system, see: https://www.bmbf.de/de/mehr-ki-erfordert-weniger-energie-8156.html <https://www.bmbf.de/de/mehr-ki-erfordert-weniger-energie-8156.html> (in German), and I am wondering what is the position of Smalltalk in that field or what it could be. From the early 90th I know that much of the pretended slowness of Smalltalk was absorbed by the more efficient programs due to better analysis, concepts, design and development tools (and maybe developers ? ). >> >> Best Regards, Bernhard > > -- > > Kind Regards, > > Sean Glazier > 603 892 0167 cell > 603 583 4575 Skype phone > Skype Id: visualwave >

April 24, 2019

Re: [Pharo-dev] FloatArray

by Jimmie Houchin

That would be awesome. There is most definitely interest. Thanks. On 4/24/19 12:20 AM, Nicolas Cellier wrote: > Hi, > I recommand inquiring about Smallapack, the Smalltalk interface to > LAPACK, on squeaksource.com <http://squeaksource.com> or github. > You'll get the speed of numpy. There is a Metacello configuration. I > have not checked the port on current > Pharo, but I can reactivate if there is some interest. > > Le mer. 24 avr. 2019 Ã 05:55, Jimmie Houchin <jlhouchin(a)gmail.com > <mailto:jlhouchin@gmail.com>> a Ã©critÂ : > > I just recently discovered FloatArray by accident. > > I was close to having to use Python/Numpy due to considerable > performance differences. Performance is very important. > > In one of my test using a 212000 item array of floats and doing a > sum on > each iteration through the array Pharo was taking 17 minutes verses > Python/Numpy taking 20 seconds. Using FloatArray closes the gap to 40 > seconds. That is acceptable. > > However when looking at the FloatArray comment it says it uses 32bit > floats. I need 64bit. I don't even know where to look to create 64bit > FloatArrays. I am surprised that they didn't get converted when > moving > to 64bit Pharo. > > Would I need to change VM source? Compile a new VM and create > image side > classes? I have not messed with VM in years. I do not know where the > FloatArray plugin would be. > > Any pointers would be a great help. If this needs to be on the vm-dev > list I can move it there. > > Thanks. > > Jimmie > >

April 24, 2019

Re: [Pharo-dev] FloatArray

by ducasse

> On 24 Apr 2019, at 07:20, Nicolas Cellier <nicolas.cellier.aka.nice(a)gmail.com> wrote: > > Hi, > I recommand inquiring about Smallapack, the Smalltalk interface to LAPACK, on squeaksource.com <http://squeaksource.com/> or github. You'll get the speed of numpy. There is a Metacello configuration. I have not checked the port on current > Pharo, but I can reactivate if there is some interest. Yes there is. I know that oleksandr wants to crunch a lot of numbers. > > Le mer. 24 avr. 2019 Ã 05:55, Jimmie Houchin <jlhouchin(a)gmail.com <mailto:jlhouchin@gmail.com>> a Ã©crit : > I just recently discovered FloatArray by accident. > > I was close to having to use Python/Numpy due to considerable > performance differences. Performance is very important. > > In one of my test using a 212000 item array of floats and doing a sum on > each iteration through the array Pharo was taking 17 minutes verses > Python/Numpy taking 20 seconds. Using FloatArray closes the gap to 40 > seconds. That is acceptable. > > However when looking at the FloatArray comment it says it uses 32bit > floats. I need 64bit. I don't even know where to look to create 64bit > FloatArrays. I am surprised that they didn't get converted when moving > to 64bit Pharo. > > Would I need to change VM source? Compile a new VM and create image side > classes? I have not messed with VM in years. I do not know where the > FloatArray plugin would be. > > Any pointers would be a great help. If this needs to be on the vm-dev > list I can move it there. > > Thanks. > > Jimmie > >

April 24, 2019

Re: [Pharo-dev] FloatArray

by Serge Stinckwich

On Wed, Apr 24, 2019 at 6:20 AM Nicolas Cellier < nicolas.cellier.aka.nice(a)gmail.com> wrote: > Hi, > I recommand inquiring about Smallapack, the Smalltalk interface to LAPACK, on > squeaksource.com or github. You'll get the speed of numpy. There is a > Metacello configuration. I have not checked the port on current > Pharo, but I can reactivate if there is some interest. > > Yes, we be nice to have a Pharo port. I would really like to integrate this with PolyMath in one way or another. Do you have a benchmark to test the performance ? We have already one implementation of SVD in PolyMath. We can compare the results with LAPACK. Not sure we can do sum of vectors or matrices with LAPACK ? Regards, -- Serge Stinckwic h Int. Research Unit on Modelling/Simulation of Complex Systems (UMMISCO) Sorbonne University (SU) French National Research Institute for Sustainable Development (IRD) U niversity of YaoundÃ© I, Cameroun "Programs must be written for people to read, and only incidentally for machines to execute." https://twitter.com/SergeStinckwich

April 24, 2019

Re: [Pharo-dev] FloatArray

by Nicolas Cellier

Hi, I recommand inquiring about Smallapack, the Smalltalk interface to LAPACK, on squeaksource.com or github. You'll get the speed of numpy. There is a Metacello configuration. I have not checked the port on current Pharo, but I can reactivate if there is some interest. Le mer. 24 avr. 2019 Ã 05:55, Jimmie Houchin <jlhouchin(a)gmail.com> a Ã©crit : > I just recently discovered FloatArray by accident. > > I was close to having to use Python/Numpy due to considerable > performance differences. Performance is very important. > > In one of my test using a 212000 item array of floats and doing a sum on > each iteration through the array Pharo was taking 17 minutes verses > Python/Numpy taking 20 seconds. Using FloatArray closes the gap to 40 > seconds. That is acceptable. > > However when looking at the FloatArray comment it says it uses 32bit > floats. I need 64bit. I don't even know where to look to create 64bit > FloatArrays. I am surprised that they didn't get converted when moving > to 64bit Pharo. > > Would I need to change VM source? Compile a new VM and create image side > classes? I have not messed with VM in years. I do not know where the > FloatArray plugin would be. > > Any pointers would be a great help. If this needs to be on the vm-dev > list I can move it there. > > Thanks. > > Jimmie > > >

April 24, 2019

FloatArray

by Jimmie Houchin

I just recently discovered FloatArray by accident. I was close to having to use Python/Numpy due to considerable performance differences. Performance is very important. In one of my test using a 212000 item array of floats and doing a sum on each iteration through the array Pharo was taking 17 minutes verses Python/Numpy taking 20 seconds. Using FloatArray closes the gap to 40 seconds. That is acceptable. However when looking at the FloatArray comment it says it uses 32bit floats. I need 64bit. I don't even know where to look to create 64bit FloatArrays. I am surprised that they didn't get converted when moving to 64bit Pharo. Would I need to change VM source? Compile a new VM and create image side classes? I have not messed with VM in years. I do not know where the FloatArray plugin would be. Any pointers would be a great help. If this needs to be on the vm-dev list I can move it there. Thanks. Jimmie

April 24, 2019

Lectures at Montreal

by StÃ©phane Ducasse

Hello Just a little mail to mention that I will visit montreal and give some lectures 23 and 24 of May. If Montreal Smalltalkers want to discuss let me know. Stef -------------------------------------------- StÃ©phane Ducasse http://stephane.ducasse.free.fr http://www.synectique.eu / http://www.pharo.org 03 59 35 87 52 Assistant: Julie Jonas FAX 03 59 57 78 50 TEL 03 59 35 86 16 S. Ducasse - Inria 40, avenue Halley, Parc Scientifique de la Haute Borne, BÃ¢t.A, Park Plaza Villeneuve d'Ascq 59650 France

April 23, 2019