From C.Klaij at marin.nl Thu Jun 4 02:29:01 2026 From: C.Klaij at marin.nl (Klaij, Christiaan) Date: Thu, 4 Jun 2026 07:29:01 +0000 Subject: [petsc-users] using intel mkl for optimized blas/lapack on amd nodes Message-ID: I came across this issue: https://urldefense.us/v3/__https://danieldk.eu/software/misc/intel-mkl-on-amd-zen__;!!G_uCfscf7eWS!fsSSG7b4cHz_fsk3H-3DI4kIopw4fIp1MW93aPTq10vZnZKOx1t7cEg2g4MbGyxZXr15OMM0jg0UU0q2p5ktYkQ$ We do configure petsc with the intel mkl for optimized blas/lapack using --with-blas-lapack-dir=${MKL_DIR}. How much would this issue affect the performance of petsc on amd nodes? Do you have the proposed workaround in place? Chris dr. ir. Christiaan Klaij | senior researcher Research & Development | CFD Development T +31 317 49 33 44 | https://urldefense.us/v3/__http://www.marin.nl__;!!G_uCfscf7eWS!fsSSG7b4cHz_fsk3H-3DI4kIopw4fIp1MW93aPTq10vZnZKOx1t7cEg2g4MbGyxZXr15OMM0jg0UU0q2h3yojEE$ -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image455803.png Type: image/png Size: 5004 bytes Desc: image455803.png URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image705130.png Type: image/png Size: 487 bytes Desc: image705130.png URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image115249.png Type: image/png Size: 504 bytes Desc: image115249.png URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image770232.png Type: image/png Size: 482 bytes Desc: image770232.png URL: From junchao.zhang at gmail.com Thu Jun 4 09:52:07 2026 From: junchao.zhang at gmail.com (Junchao Zhang) Date: Thu, 4 Jun 2026 09:52:07 -0500 Subject: [petsc-users] using intel mkl for optimized blas/lapack on amd nodes In-Reply-To: References: Message-ID: Richard, do you want to trick Intel? --Junchao Zhang On Thu, Jun 4, 2026 at 2:29?AM Klaij, Christiaan via petsc-users < petsc-users at mcs.anl.gov> wrote: > I came across this issue: > > https://urldefense.us/v3/__https://danieldk.eu/software/misc/intel-mkl-on-amd-zen__;!!G_uCfscf7eWS!dajR2MuTfP2rxfSl3EfNQWNARhE0YDaEkG1Oa1g3PU3nT34viNVZXfsr3ssEuPluCFsljOlWzidoiyoX9ra19wA3XYtS$ > > > We do configure petsc with the intel mkl for optimized > blas/lapack using --with-blas-lapack-dir=${MKL_DIR}. > > How much would this issue affect the performance of petsc on amd > nodes? Do you have the proposed workaround in place? > > Chris > ???? > dr. ir. Christiaan Klaij | senior researcher > Research & Development | CFD Development > T +31 317 49 33 44 <+31%20317%2049%2033%2044> | https://urldefense.us/v3/__http://www.marin.nl__;!!G_uCfscf7eWS!dajR2MuTfP2rxfSl3EfNQWNARhE0YDaEkG1Oa1g3PU3nT34viNVZXfsr3ssEuPluCFsljOlWzidoiyoX9ra19770ELRq$ > > [image: Facebook] > > [image: LinkedIn] > > [image: YouTube] > > -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image455803.png Type: image/png Size: 5004 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image705130.png Type: image/png Size: 487 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image115249.png Type: image/png Size: 504 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image770232.png Type: image/png Size: 482 bytes Desc: not available URL: From pierre at joliv.et Thu Jun 4 10:03:28 2026 From: pierre at joliv.et (Pierre Jolivet) Date: Thu, 4 Jun 2026 17:03:28 +0200 Subject: [petsc-users] using intel mkl for optimized blas/lapack on amd nodes In-Reply-To: References: Message-ID: <88928E41-CBE8-4BA9-938A-F2FAD6972A7E@joliv.et> The libfakeintel.so trick is as old as I can remember and it is available in all computing centers I?m using. But it?s the duty of the users to preload it (again, on computing centers I have access to, module load?ing the MKL on AMD nodes automatically sets LD_PRELOAD). Thanks, Pierre > On 4 Jun 2026, at 4:52?PM, Junchao Zhang wrote: > > Richard, do you want to trick Intel? > > --Junchao Zhang > > > On Thu, Jun 4, 2026 at 2:29?AM Klaij, Christiaan via petsc-users > wrote: >> I came across this issue: >> >> https://urldefense.us/v3/__https://danieldk.eu/software/misc/intel-mkl-on-amd-zen__;!!G_uCfscf7eWS!bSuPjhB4YML6KMVUaPPLNST5KPen-7_1fwZ4ZQOjJRo4oxNXqPVG-kHukhNSni7bEbvwnpMNfHLAs7Dwqqckeg$ >> >> We do configure petsc with the intel mkl for optimized >> blas/lapack using --with-blas-lapack-dir=${MKL_DIR}. >> >> How much would this issue affect the performance of petsc on amd >> nodes? Do you have the proposed workaround in place? >> >> Chris >> >> ???? >> dr. ir. Christiaan Klaij | senior researcher >> Research & Development | CFD Development >> T +31?317?49?33?44 | https://urldefense.us/v3/__http://www.marin.nl__;!!G_uCfscf7eWS!bSuPjhB4YML6KMVUaPPLNST5KPen-7_1fwZ4ZQOjJRo4oxNXqPVG-kHukhNSni7bEbvwnpMNfHLAs7BwifXosQ$ >> >> >> -------------- next part -------------- An HTML attachment was scrubbed... URL: From junchao.zhang at gmail.com Thu Jun 4 11:13:27 2026 From: junchao.zhang at gmail.com (Junchao Zhang) Date: Thu, 4 Jun 2026 11:13:27 -0500 Subject: [petsc-users] using intel mkl for optimized blas/lapack on amd nodes In-Reply-To: <88928E41-CBE8-4BA9-938A-F2FAD6972A7E@joliv.et> References: <88928E41-CBE8-4BA9-938A-F2FAD6972A7E@joliv.et> Message-ID: Can we add the fake function to petsc source code? --Junchao Zhang On Thu, Jun 4, 2026 at 10:04?AM Pierre Jolivet wrote: > The libfakeintel.so trick is as old as I can remember and it is available > in all computing centers I?m using. > But it?s the duty of the users to preload it (again, on computing centers > I have access to, module load?ing the MKL on AMD nodes automatically sets > LD_PRELOAD). > > Thanks, > Pierre > > On 4 Jun 2026, at 4:52?PM, Junchao Zhang wrote: > > Richard, do you want to trick Intel? > > --Junchao Zhang > > > On Thu, Jun 4, 2026 at 2:29?AM Klaij, Christiaan via petsc-users < > petsc-users at mcs.anl.gov> wrote: > >> I came across this issue: >> >> https://urldefense.us/v3/__https://danieldk.eu/software/misc/intel-mkl-on-amd-zen__;!!G_uCfscf7eWS!aISYg-KogJlnj88A8WDaRhFjww0fxMixLU0FERmdocFSoqjBBWJ8d-rxkrD3ZogxTqqYgUp47LMQCrFtNVnSQnsTWpZ5$ >> >> >> We do configure petsc with the intel mkl for optimized >> blas/lapack using --with-blas-lapack-dir=${MKL_DIR}. >> >> How much would this issue affect the performance of petsc on amd >> nodes? Do you have the proposed workaround in place? >> >> Chris >> ???? >> dr. ir. Christiaan Klaij | senior researcher >> Research & Development | CFD Development >> T +31 317 49 33 44 <+31%20317%2049%2033%2044> | https://urldefense.us/v3/__http://www.marin.nl__;!!G_uCfscf7eWS!aISYg-KogJlnj88A8WDaRhFjww0fxMixLU0FERmdocFSoqjBBWJ8d-rxkrD3ZogxTqqYgUp47LMQCrFtNVnSQpOuyUpE$ >> >> >> >> >> >> >> >> > > -------------- next part -------------- An HTML attachment was scrubbed... URL: From balay.anl at fastmail.org Thu Jun 4 11:19:50 2026 From: balay.anl at fastmail.org (Satish Balay) Date: Thu, 4 Jun 2026 11:19:50 -0500 (CDT) Subject: [petsc-users] using intel mkl for optimized blas/lapack on amd nodes In-Reply-To: References: Message-ID: <50e622d9-f669-28db-a22e-3033bafb9864@fastmail.org> Hm PETSc primarily uses level1 blas - where the performance differences are likely not huge. So these workarounds might not be worth the effort [vs using AMD blas]? https://urldefense.us/v3/__https://www.amd.com/en/developer/aocl/blis.html__;!!G_uCfscf7eWS!ZlLZSYSJ2CPe8ib18GTsikw1SdMg_bSjP6PwhGwHErFBZzYvLwjVBzGFSpktQDr6AQfkO6GKO6YG8Blb0CUxflllj2s$ Satish On Thu, 4 Jun 2026, Klaij, Christiaan via petsc-users wrote: > I came across this issue: > > https://urldefense.us/v3/__https://danieldk.eu/software/misc/intel-mkl-on-amd-zen__;!!G_uCfscf7eWS!fsSSG7b4cHz_fsk3H-3DI4kIopw4fIp1MW93aPTq10vZnZKOx1t7cEg2g4MbGyxZXr15OMM0jg0UU0q2p5ktYkQ$ > > We do configure petsc with the intel mkl for optimized > blas/lapack using --with-blas-lapack-dir=${MKL_DIR}. > > How much would this issue affect the performance of petsc on amd > nodes? Do you have the proposed workaround in place? > > Chris > dr. ir. Christiaan Klaij | senior researcher > Research & Development | CFD Development > T +31 317 49 33 44 | https://urldefense.us/v3/__http://www.marin.nl__;!!G_uCfscf7eWS!fsSSG7b4cHz_fsk3H-3DI4kIopw4fIp1MW93aPTq10vZnZKOx1t7cEg2g4MbGyxZXr15OMM0jg0UU0q2h3yojEE$ > From stefano.zampini at gmail.com Thu Jun 4 11:35:48 2026 From: stefano.zampini at gmail.com (Stefano Zampini) Date: Thu, 4 Jun 2026 17:35:48 +0100 Subject: [petsc-users] using intel mkl for optimized blas/lapack on amd nodes In-Reply-To: <50e622d9-f669-28db-a22e-3033bafb9864@fastmail.org> References: <50e622d9-f669-28db-a22e-3033bafb9864@fastmail.org> Message-ID: Given that the fix uses LD_PRELOAD, I don't think it should be our responsibility Il giorno gio 4 giu 2026 alle ore 17:20 Satish Balay ha scritto: > Hm PETSc primarily uses level1 blas - where the performance differences > are likely not huge. > > So these workarounds might not be worth the effort [vs using AMD blas]? > > > https://urldefense.us/v3/__https://www.amd.com/en/developer/aocl/blis.html__;!!G_uCfscf7eWS!ZlLZSYSJ2CPe8ib18GTsikw1SdMg_bSjP6PwhGwHErFBZzYvLwjVBzGFSpktQDr6AQfkO6GKO6YG8Blb0CUxflllj2s$ > > Satish > > On Thu, 4 Jun 2026, Klaij, Christiaan via petsc-users wrote: > > > I came across this issue: > > > > > https://urldefense.us/v3/__https://danieldk.eu/software/misc/intel-mkl-on-amd-zen__;!!G_uCfscf7eWS!fsSSG7b4cHz_fsk3H-3DI4kIopw4fIp1MW93aPTq10vZnZKOx1t7cEg2g4MbGyxZXr15OMM0jg0UU0q2p5ktYkQ$ > > > > We do configure petsc with the intel mkl for optimized > > blas/lapack using --with-blas-lapack-dir=${MKL_DIR}. > > > > How much would this issue affect the performance of petsc on amd > > nodes? Do you have the proposed workaround in place? > > > > Chris > > dr. ir. Christiaan Klaij | senior researcher > > Research & Development | CFD Development > > T +31 317 49 33 44 | > https://urldefense.us/v3/__http://www.marin.nl__;!!G_uCfscf7eWS!fsSSG7b4cHz_fsk3H-3DI4kIopw4fIp1MW93aPTq10vZnZKOx1t7cEg2g4MbGyxZXr15OMM0jg0UU0q2h3yojEE$ > > > > -- Stefano -------------- next part -------------- An HTML attachment was scrubbed... URL: From pierre at joliv.et Thu Jun 4 12:54:17 2026 From: pierre at joliv.et (Pierre Jolivet) Date: Thu, 4 Jun 2026 19:54:17 +0200 Subject: [petsc-users] using intel mkl for optimized blas/lapack on amd nodes In-Reply-To: <50e622d9-f669-28db-a22e-3033bafb9864@fastmail.org> References: <50e622d9-f669-28db-a22e-3033bafb9864@fastmail.org> Message-ID: > On 4 Jun 2026, at 6:20?PM, Satish Balay wrote: > > ?Hm PETSc primarily uses level1 blas - where the performance differences are likely not huge. > > So these workarounds might not be worth the effort [vs using AMD blas]? They are definitely worth the effort for direct solvers which use BLAS 3 extensively. But again, this is in the hand of the users, IMO. Pierre > > https://urldefense.us/v3/__https://www.amd.com/en/developer/aocl/blis.html__;!!G_uCfscf7eWS!ZlLZSYSJ2CPe8ib18GTsikw1SdMg_bSjP6PwhGwHErFBZzYvLwjVBzGFSpktQDr6AQfkO6GKO6YG8Blb0CUxflllj2s$ > > Satish > >> On Thu, 4 Jun 2026, Klaij, Christiaan via petsc-users wrote: >> >> I came across this issue: >> >> https://urldefense.us/v3/__https://danieldk.eu/software/misc/intel-mkl-on-amd-zen__;!!G_uCfscf7eWS!fsSSG7b4cHz_fsk3H-3DI4kIopw4fIp1MW93aPTq10vZnZKOx1t7cEg2g4MbGyxZXr15OMM0jg0UU0q2p5ktYkQ$ >> >> We do configure petsc with the intel mkl for optimized >> blas/lapack using --with-blas-lapack-dir=${MKL_DIR}. >> >> How much would this issue affect the performance of petsc on amd >> nodes? Do you have the proposed workaround in place? >> >> Chris >> dr. ir. Christiaan Klaij | senior researcher >> Research & Development | CFD Development >> T +31 317 49 33 44 | https://urldefense.us/v3/__http://www.marin.nl__;!!G_uCfscf7eWS!fsSSG7b4cHz_fsk3H-3DI4kIopw4fIp1MW93aPTq10vZnZKOx1t7cEg2g4MbGyxZXr15OMM0jg0UU0q2h3yojEE$ >> > From C.Klaij at marin.nl Fri Jun 5 02:07:40 2026 From: C.Klaij at marin.nl (Klaij, Christiaan) Date: Fri, 5 Jun 2026 07:07:40 +0000 Subject: [petsc-users] using intel mkl for optimized blas/lapack on amd nodes In-Reply-To: <88928E41-CBE8-4BA9-938A-F2FAD6972A7E@joliv.et> References: <88928E41-CBE8-4BA9-938A-F2FAD6972A7E@joliv.et> Message-ID: If the users are supposed to know this and act on it, perhaps an entry in the FAQs to raise awareness? With instruction how to check if the user's computing center automatically applies the trick. Chris _____ dr. ir. Christiaan Klaij | senior researcher Research & Development | CFD Development T +31 317 49 33 44 | https://urldefense.us/v3/__http://www.marin.nl__;!!G_uCfscf7eWS!bxo9uS0gHxRz9KO1sRNC2XOSjsx85ZEUIJkT1EPc6PG9lqapUvessDpxuTI0IWBGYwe8Xj5omRetklwCifJyIqw$ ___________________________________ From: Pierre Jolivet Sent: Thursday, June 4, 2026 5:03 PM To: Junchao Zhang Cc: Richard Tran Mills; PETSc users list; Klaij, Christiaan Subject: Re: [petsc-users] using intel mkl for optimized blas/lapack on amd nodes You don't often get email from pierre at joliv.et. Learn why this is important The libfakeintel.so trick is as old as I can remember and it is available in all computing centers I?m using. But it?s the duty of the users to preload it (again, on computing centers I have access to, module load?ing the MKL on AMD nodes automatically sets LD_PRELOAD). Thanks, Pierre On 4 Jun 2026, at 4:52?PM, Junchao Zhang wrote: Richard, do you want to trick Intel? --Junchao Zhang On Thu, Jun 4, 2026 at 2:29?AM Klaij, Christiaan via petsc-users > wrote: I came across this issue: https://urldefense.us/v3/__https://danieldk.eu/software/misc/intel-mkl-on-amd-zen__;!!G_uCfscf7eWS!bxo9uS0gHxRz9KO1sRNC2XOSjsx85ZEUIJkT1EPc6PG9lqapUvessDpxuTI0IWBGYwe8Xj5omRetklwCbUAVriM$ We do configure petsc with the intel mkl for optimized blas/lapack using --with-blas-lapack-dir=${MKL_DIR}. How much would this issue affect the performance of petsc on amd nodes? Do you have the proposed workaround in place? Chris ???? dr. ir. Christiaan Klaij | senior researcher Research & Development | CFD Development T +31 317 49 33 44 | https://urldefense.us/v3/__http://www.marin.nl__;!!G_uCfscf7eWS!bxo9uS0gHxRz9KO1sRNC2XOSjsx85ZEUIJkT1EPc6PG9lqapUvessDpxuTI0IWBGYwe8Xj5omRetklwCifJyIqw$ -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image715333.png Type: image/png Size: 5004 bytes Desc: image715333.png URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image900709.png Type: image/png Size: 487 bytes Desc: image900709.png URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image751687.png Type: image/png Size: 504 bytes Desc: image751687.png URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image579641.png Type: image/png Size: 482 bytes Desc: image579641.png URL: From daniele.prada85 at gmail.com Mon Jun 8 11:30:05 2026 From: daniele.prada85 at gmail.com (Daniele Prada) Date: Mon, 8 Jun 2026 18:30:05 +0200 Subject: [petsc-users] DMPlexRestoreCellCoordinates failing with optimized PETSc mode Message-ID: Hello everyone, I am having the following message: [0]PETSC ERROR: Caught signal number 11 SEGV: Segmentation Violation, probably memory access out of range when using DMPlexRestoreCellCoordinates() with PETSc built in optimized mode, whereas the same function is not giving any trouble with PETSc built in debug mode. By looking into the implementation of DMPlexRestoreCellCoordinates(), it seems the issue comes from DMPlexVecGetOrientedClosure(). The debug mode has been built with: ./configure PETSC_ARCH=arch-linux-c-debug --download-fblaslapack --download-mpich --with-64-bit-integers --with-strict-petscerrorcode whereas the optimized one with: ./configure PETSC_ARCH=arch-linux-c-opt --with-debugging=0 COPTFLAGS="-O3 -march=native -mtune=native" CXXOPTFLAGS="-O3 -march=native -mtune=native" FOPTFLAGS="-O3 -march=native -mtune=native" --download-fblaslapack --download-mpich Other info: - I am working with an Intel Xeon laptop - uname -a: Linux daniele-Precision-7560 6.17.0-35-generic #35~24.04.1-Ubuntu SMP PREEMPT_DYNAMIC Tue May 26 19:30:42 UTC 2 x86_64 x86_64 x86_64 GNU/Linux - PETSc version: the most recent one - Compiler: gcc 13.3 Attached, please find: - ex03.c: a selection of the relevant lines of the main program that are causing the issue. Note: DMPlexCreateFromOVMFile() is a function I wrote to create a fully interpolated (and symmetric) DMPlex from a .ovm file - torus_from_CVT2d_Npol_000624.ovm: the .ovm file containing a 3D mesh. Thank you in advance for your assistance Best regards Daniele Prada -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ex03.c Type: text/x-csrc Size: 2031 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: torus_from_CVT2d_Npol_000624.ovm Type: application/octet-stream Size: 157107 bytes Desc: not available URL: From junchao.zhang at gmail.com Mon Jun 8 12:12:53 2026 From: junchao.zhang at gmail.com (Junchao Zhang) Date: Mon, 8 Jun 2026 12:12:53 -0500 Subject: [petsc-users] using intel mkl for optimized blas/lapack on amd nodes In-Reply-To: References: <88928E41-CBE8-4BA9-938A-F2FAD6972A7E@joliv.et> Message-ID: Hi, Chris, Could you check out this MR https://urldefense.us/v3/__https://gitlab.com/petsc/petsc/-/merge_requests/9327__;!!G_uCfscf7eWS!ctF3nShH_6dTwbShHGEKUtwTmiwcG7QifVwCzUXYkhfzZUUbY2Qjzn9hPfKSa3jOWkC94o2dELHf38z0XsL4xh-DrwXl$ and test the attached ex00.c in your environment to see if it makes a difference? On my machine, it helps a little, as the MKL version alway optimizes for Zen. Thanks! --Junchao Zhang On Fri, Jun 5, 2026 at 2:07?AM Klaij, Christiaan wrote: > If the users are supposed to know this and act on it, perhaps an entry in > the FAQs to raise awareness? With instruction how to check if the user's > computing center automatically applies the trick. > > Chris > > ________________________________________ > ???? > dr. ir. Christiaan Klaij | senior researcher > Research & Development | CFD Development > T +31 317 49 33 44 <+31%20317%2049%2033%2044> | https://urldefense.us/v3/__http://www.marin.nl__;!!G_uCfscf7eWS!ctF3nShH_6dTwbShHGEKUtwTmiwcG7QifVwCzUXYkhfzZUUbY2Qjzn9hPfKSa3jOWkC94o2dELHf38z0XsL4xgaQGxNe$ > [image: Facebook] > [image: LinkedIn] > [image: YouTube] > > From: Pierre Jolivet > Sent: Thursday, June 4, 2026 5:03 PM > To: Junchao Zhang > Cc: Richard Tran Mills; PETSc users list; Klaij, Christiaan > Subject: Re: [petsc-users] using intel mkl for optimized blas/lapack on > amd nodes > > You don't often get email from pierre at joliv.et. Learn why this is > important > The libfakeintel.so trick is as old as I can > remember and it is available in all computing centers I?m using. > But it?s the duty of the users to preload it (again, on computing centers > I have access to, module load?ing the MKL on AMD nodes automatically sets > LD_PRELOAD). > > Thanks, > Pierre > > On 4 Jun 2026, at 4:52?PM, Junchao Zhang wrote: > > Richard, do you want to trick Intel? > > --Junchao Zhang > > > On Thu, Jun 4, 2026 at 2:29?AM Klaij, Christiaan via petsc-users < > petsc-users at mcs.anl.gov> wrote: > I came across this issue: > > https://urldefense.us/v3/__https://danieldk.eu/software/misc/intel-mkl-on-amd-zen__;!!G_uCfscf7eWS!ctF3nShH_6dTwbShHGEKUtwTmiwcG7QifVwCzUXYkhfzZUUbY2Qjzn9hPfKSa3jOWkC94o2dELHf38z0XsL4xlBmyDlt$ < > https://urldefense.us/v3/__https://danieldk.eu/software/misc/intel-mkl-on-amd-zen__;!!G_uCfscf7eWS!fsSSG7b4cHz_fsk3H-3DI4kIopw4fIp1MW93aPTq10vZnZKOx1t7cEg2g4MbGyxZXr15OMM0jg0UU0q2p5ktYkQ$ > > > > We do configure petsc with the intel mkl for optimized > blas/lapack using --with-blas-lapack-dir=${MKL_DIR}. > > How much would this issue affect the performance of petsc on amd > nodes? Do you have the proposed workaround in place? > > Chris > > ???? > dr. ir. Christiaan Klaij | senior researcher > Research & Development | CFD Development > T +31 317 49 33 44 | https://urldefense.us/v3/__http://www.marin.nl__;!!G_uCfscf7eWS!ctF3nShH_6dTwbShHGEKUtwTmiwcG7QifVwCzUXYkhfzZUUbY2Qjzn9hPfKSa3jOWkC94o2dELHf38z0XsL4xgaQGxNe$ < > https://urldefense.us/v3/__https://www.marin.nl/__;!!G_uCfscf7eWS!fsSSG7b4cHz_fsk3H-3DI4kIopw4fIp1MW93aPTq10vZnZKOx1t7cEg2g4MbGyxZXr15OMM0jg0UU0q2Ae8qhHs$ > > > < > https://urldefense.us/v3/__https://www.facebook.com/marin.wageningen__;!!G_uCfscf7eWS!fsSSG7b4cHz_fsk3H-3DI4kIopw4fIp1MW93aPTq10vZnZKOx1t7cEg2g4MbGyxZXr15OMM0jg0UU0q2P24LxDM$ > > > < > https://urldefense.us/v3/__https://www.linkedin.com/company/marin__;!!G_uCfscf7eWS!fsSSG7b4cHz_fsk3H-3DI4kIopw4fIp1MW93aPTq10vZnZKOx1t7cEg2g4MbGyxZXr15OMM0jg0UU0q2D0cffrY$ > > > < > https://urldefense.us/v3/__https://www.youtube.com/marinmultimedia__;!!G_uCfscf7eWS!fsSSG7b4cHz_fsk3H-3DI4kIopw4fIp1MW93aPTq10vZnZKOx1t7cEg2g4MbGyxZXr15OMM0jg0UU0q2xfPbYxc$ > > > > -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image715333.png Type: image/png Size: 5004 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image900709.png Type: image/png Size: 487 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image751687.png Type: image/png Size: 504 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image579641.png Type: image/png Size: 482 bytes Desc: not available URL: From rpoli at utexas.edu Fri Jun 12 17:04:30 2026 From: rpoli at utexas.edu (Renato E Poli) Date: Fri, 12 Jun 2026 17:04:30 -0500 Subject: [petsc-users] Using CPR with PETSC and GPU Message-ID: Hi We are trying to use a CPR preconditioner for a multiphase flow solver using the fieldsplit option. It seems to work. We see a reduced number of iterations. However, it is not using the GPU effectively. Is there any documentation that could help? Is the fieldsplit+GPU supported? (v. 3.25.2) These are the options we are using (after a bit of trial-and-error): -ksp_type fgmres -ksp_gmres_restart 40 -pc_fieldsplit_type multiplicative -fieldsplit_pressure_ksp_type preonly -fieldsplit_pressure_pc_type hypre -fieldsplit_pressure_pc_hypre_type boomeramg -fieldsplit_other_ksp_type gmres -fieldsplit_other_ksp_gmres_restart 5 -fieldsplit_other_ksp_max_it 5 -fieldsplit_other_ksp_rtol 1e-1 -fieldsplit_other_pc_type bjacobi -fieldsplit_other_sub_pc_type ilu -fieldsplit_other_sub_pc_factor_reuse_ordering true -fieldsplit_other_sub_pc_factor_reuse_fill true Thank you, Renato -------------- next part -------------- An HTML attachment was scrubbed... URL: