October 2023 - Ecp-aurora-sdk-discuss

Re: [Ecp-aurora-sdk-discuss] New oneAPI release and eng-compiler versions
by Bertoni, Colleen 25 Oct '23

25 Oct '23

Thanks Chris! Comparing this new release to the current latest (oneapi/eng-compiler/2023.05.15.007) we see 25 fixes and 7 regressions in the bug reproducers (below). Several of the regressions should be fixed with the 682 UMD, which is on Aurora. The big change to keep in mind, like Chris mentioned, is the new directory structure. So if you were hardcoding to a library at MKLROOT/lib/intel64 for example, you’ll need to change it. One other thing to keep in mind is that if you were using the subdevice() clause in OpenMP to target GPU tiles explicitly, that has been deprecated. Instead you can use gpu_tile_compact.sh or ONEAPI_DEVICE_SELECTOR to target tiles. Let us know if you have any questions, and thanks for all the bug reporting! Colleen Fixes: > ./ath.py compare --list --status=fixed --reference-file standard/2023.05.15.007/reproduc\ ers/output.json --new-file standard/2023.10.15.002-rc03/reproducers/output.json source/reproducers/dpcpp/ms371-InlineAsm:ms371-InlineAsm_asm_bad_opcode [cmplrllvm-46097] source/reproducers/dpcpp/ms371-InlineAsm:ms371-InlineAsm_asm_bad_operand_syntax [cmplrllvm-46097] source/reproducers/dpcpp/ms371-InlineAsm:ms371-InlineAsm_asm_duplicate_label [cmplrllvm-46097] source/reproducers/dpcpp/ms371-InlineAsm:ms371-InlineAsm_asm_illegal_exec_size [cmplrllvm-46097] source/reproducers/dpcpp/ms371-InlineAsm:ms371-InlineAsm_asm_missing_label [cmplrllvm-46097] source/reproducers/dpcpp/ms371-InlineAsm:ms371-InlineAsm_asm_missing_region [cmplrllvm-46097] source/reproducers/dpcpp/ms371-InlineAsm:ms371-InlineAsm_asm_simple [cmplrllvm-46097] source/reproducers/dpcpp/ms371-InlineAsm:ms371-InlineAsm_asm_undefined_decl [cmplrllvm-46097] source/reproducers/dpcpp/ms371-InlineAsm:ms371-InlineAsm_asm_undefined_pred [cmplrllvm-46097] source/reproducers/dpcpp/ms371-InlineAsm:ms371-InlineAsm_asm_wrong_declare [cmplrllvm-46097] source/reproducers/dpcpp/zedebug_issue [GSD-4854] source/reproducers/ifx/dcmesh_io source/reproducers/mkl/SYCL_USM_MKL_L0_subdevice:SYCL_USM_MKL_L0_subdevice_10 [CMPLRLLVM-39495,MKLD-15722,cmplrllvm-48893] source/reproducers/mkl/SYCL_USM_MKL_L0_subdevice:SYCL_USM_MKL_L0_subdevice_11 [CMPLRLLVM-39495,MKLD-15722,cmplrllvm-48893] source/reproducers/mkl/gemv_beta_zero:gemv_beta_zero_ath_run source/reproducers/mkl/gemv_beta_zero:gemv_beta_zero_fortran source/reproducers/mkl/potrf_debug:potrf_debug_ath_run source/reproducers/mkl/zgetrs_batch_slowdown [MKLD-15212,MKLD-15906] source/reproducers/openmp/fortran_link [CMPLRLLVM-48607] source/reproducers/openmp/fortran_private_block:fortran_private_block_private_local_block [CMPLRLLVM-49146] P0 source/reproducers/openmp/fortran_slice_leak source/reproducers/openmp/oneapi_device_selector_test:oneapi_device_selector_test_ath_run source/reproducers/tools/advisor_gflop:advisor_gflop.w_a_DPCPP_app source/reproducers/tools/vtune_amrex_vector:vtune_amrex_vector.amrex_vector_vtune source/reproducers/tools/vtune_jit_openmp [VASP-29284] Regressions: > ./ath.py compare --list --status=regressions --reference-file standard/2023.05.15.007/re\producers/output.json --new-file standard/2023.10.15.002-rc03/reproducers/output.json source/reproducers/dpcpp/ATSsubdevice_regression # hanging, needs new UMD source/reproducers/mkl/minimal_zgeqrf_tester:minimal_zgeqrf_tester_bigger_ze00_ath_run [mkld-15938] source/reproducers/mkl/minimal_zgeqrf_tester:minimal_zgeqrf_tester_ze00_ath_run [mkld-15938] source/reproducers/mkl/minimal_zgeqrf_tester:minimal_zgeqrf_tester_ze0_ath_run [mkld-15938] source/reproducers/openmp/findclus:findclus_ath_run [GSD-5335] # needs newer KMD source/reproducers/openmp/gamess_mini_all_kernels:gamess_mini_all_kernels_j16 [CMPLRLLVM-45082,CMPLRLLVM-45890,CMPLRTST-21020,IGC-6246,XDEPS-5609] source/reproducers/openmp/gamess_mini_all_kernels:gamess_mini_all_kernels_j19 [CMPLRLLVM-45082,CMPLRLLVM-45890,CMPLRTST-21020,IGC-6246,XDEPS-5609] # needs newer KMD/UMD ________________________________ From: Ecp-aurora-sdk-discuss <ecp-aurora-sdk-discuss-bounces(a)lists.jlse.anl.gov> on behalf of Chan-nui, Christopher via Ecp-aurora-sdk-discuss <ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov> Sent: Tuesday, October 24, 2023 5:00 PM To: ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov <ecp-aurora-sdk-discuss(a)lists.jlse.anl.gov> Subject: [Ecp-aurora-sdk-discuss] New oneAPI release and eng-compiler versions A new version of the Aurora SDK is now available. This will NOT be enabled as the default version. Running jobs will NOT be affected. The module files for this release are: oneapi/release/2023.10.15.001 oneapi/eng-compiler/2023.10.15.002 This is based on the 2023.2.0 release of the oneAPI SDK. Release notes are available at https://www.intel.com/content/www/us/en/developer/articles/release-notes/in… https://www.intel.com/content/www/us/en/developer/articles/release-notes/in… Addendums for the engineering components are available at https://wiki.jlse.anl.gov/display/inteldga/DGA+SDK+Release+Status+and+Notes. # IMPORTANT NOTE: The directory structure for engineering components has changed. The directory layout has been flattend and the 64 bit executables/libraries are now the default. Individual files may have been relocated as well. Notably: For the compiler ${CMPROOT}/bin/intel64 -> ${CMPLR_ROOT}/bin ${CMPROOT}/compiler/lib/intel64_lin -> ${CMPLR_ROOT}/lib For MKL ${MKL_ROOT}/bin/intel64 -> ${MKL_ROOT}/bin ${MKL_ROOT}/lib/intel64 -> ${MKL_ROOT}/lib ${MKL_ROOT}/examples -> ${MKL_ROOT}/share/doc/mkl/examples Colleen will provide caveats, workarounds and other useful information about these releases shortly. -- Christopher Chan-Nui christopher.chan-nui(a)intel.com

1 0

DL Frameworks 2023.2 release available
by Ramer, Rebecca 24 Oct '23

24 Oct '23

Hi, I would like to announce availability of the 2023.2 DL Frameworks release with support for pytorch and tensorflow on the Aurora and Sunspot deployments. This release is based on the 2023.2 versions of the Deep Learning Frameworks and uses the recently announced oneAPI SDK release based on the Intel 2023.2.0 oneapi basekit via the oneapi/release/2023.10.15.001 modulefile. This release is available by loading the frameworks modulefile: module load frameworks/2023.10.15.001 Loading this module activates the aurora_models_frameworks-2023.2 conda environment. Along with the pytorch and tensorflow packages, all necessary python packages needed to run the 11 'Aurora NRE' models have been installed into this conda environment. This frameworks modulefile replaces the 'hidden' modulefile frameworks/.2023.08.15.002 and is the default frameworks modulefile. All NRE model workloads have been validated using the frameworks/2023.10.15.001 modulefile. No new issues were encountered. Validation on Aurora will be done when it becomes available to use for Aurora NRE model workloads. More information about the 2023.2 DL Frameworks release can be found here: https://wiki.jlse.anl.gov/display/inteldga/DL+Frameworks+on+Sunspot Please let me know of any issues encountered. Rebecca Ramer rebecca.ramer(a)intel.com<mailto:[email protected]>

1 0

New oneAPI release and eng-compiler versions
by Chan-nui, Christopher 24 Oct '23

24 Oct '23

A new version of the Aurora SDK is now available. This will NOT be enabled as the default version. Running jobs will NOT be affected. The module files for this release are: oneapi/release/2023.10.15.001 oneapi/eng-compiler/2023.10.15.002 This is based on the 2023.2.0 release of the oneAPI SDK. Release notes are available at https://www.intel.com/content/www/us/en/developer/articles/release-notes/in… https://www.intel.com/content/www/us/en/developer/articles/release-notes/in… Addendums for the engineering components are available at https://wiki.jlse.anl.gov/display/inteldga/DGA+SDK+Release+Status+and+Notes. # IMPORTANT NOTE: The directory structure for engineering components has changed. The directory layout has been flattend and the 64 bit executables/libraries are now the default. Individual files may have been relocated as well. Notably: For the compiler ${CMPROOT}/bin/intel64 -> ${CMPLR_ROOT}/bin ${CMPROOT}/compiler/lib/intel64_lin -> ${CMPLR_ROOT}/lib For MKL ${MKL_ROOT}/bin/intel64 -> ${MKL_ROOT}/bin ${MKL_ROOT}/lib/intel64 -> ${MKL_ROOT}/lib ${MKL_ROOT}/examples -> ${MKL_ROOT}/share/doc/mkl/examples Colleen will provide caveats, workarounds and other useful information about these releases shortly. -- Christopher Chan-Nui christopher.chan-nui(a)intel.com

1 0