QUDA with HIP:rocm commitshttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commits/rocm2019-11-18T10:28:20+08:00https://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/2a5b7e4a6d8fef849622e900fa371e02f3e8f5b6Update README.md2019-11-18T10:28:20+08:00YI-Bo Yangybyang@itp.ac.cnhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/a6772342534099eb0dc346acba13216240550885Add an example of Makefile to setup cmake and make quda.2019-11-09T23:16:02+08:00Yi-Boybyang@itp.ac.cnhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/28f6d51eed9301060026c00bdb7a0eba5e9fa9f5Generate the *.make through FindHip.2019-11-09T22:57:52+08:00Yi-Boybyang@itp.ac.cnhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/0c4590e7c815f8693d0dc62c3aeb6e8824e0adf0Hacks for the bugs of present HIPCC:2019-11-09T12:44:09+08:00Yi-Boybyang@itp.ac.cn
1. hipmemcpyToSymbol doesn't work.
2. Use certain argument classs on the host through __global__ xx(Arg arg)
can get unexpected result.
Both the hacks can reduce the performance and can be discarded when the bugs
are fixed.https://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/98be40f5ae825204cd9dad74d1ed810adb4de456Hack on the memset2D as the illegal address can be requested in the...2019-11-08T20:38:06+08:00Yi-Boybyang@itp.ac.cnHack on the memset2D as the illegal address can be requested in the hipmemeset2D, even though it will not be touched.
https://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/b3189c4ac19c805268bc252987042a996508c3c0hack on the dslash launcher to avoid quda_launch_kernel which can not be used...2019-11-08T20:35:51+08:00Yi-Boybyang@itp.ac.cnhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/02c21800b8f18b98f3b2fc914037a376d3598bd4Hacks on the maximum threads used in the tuning to aovid crashes.2019-11-08T20:34:51+08:00Yi-Boybyang@itp.ac.cnhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/9649269f452e7f6375fc4d01f7d7aaeec08e5e7fMake the error message to be more precise.2019-11-08T20:34:15+08:00Yi-Boybyang@itp.ac.cnhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/7664f498b2314caf8406d1f6fc32d6b1f0c93e1eHack on the pointer check as HIP can not recognize a host pointer2019-11-08T20:32:31+08:00Yi-Boybyang@itp.ac.cnhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/f5ce07a9538acae57f4e2153576b78490ac9c3a2Hacks on the random number generator as it doesn't work.2019-11-08T20:31:33+08:00Yi-Boybyang@itp.ac.cnhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/966c16048a2f8fcc2fb4082104198caef17b6eb7Rename thrust::cuda::par into thrust_par to avoid the direct dependence.2019-11-08T20:28:41+08:00Yi-Boybyang@itp.ac.cnhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/fe729b179fd2691a97ef6de88ef8076a859e7fa9The version can be compiled2019-11-04T04:18:26+08:00Yi-Boybyang@itp.ac.cn
The c++ compiler is set to be hipcc. Thus the cmake should be hacked to include .cu in the suffix can be indentified to be the cppfiles. Otherwise one can rename all the .cu files into .cc to avoid the hack on cmake.https://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/9505d29fec47f1f1e3ff424545562217e599bb1eadd a local copy of thrust which works with quda2019-11-03T16:34:53+08:00Yi-Boybyang@itp.ac.cnhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/1d6f8443add365ce72788ab5a71bbad15c66635cChanges on the Cmake files2019-11-03T15:47:08+08:00Yi-Boybyang@itp.ac.cn
please configure quda with following command (CMAKE_BUILD_TYPE DEVEL is hacked. Note that -O3 is necessary and -g should be avoided)
MPICXX=mpicxx MPICC=mpicc CC=gcc CXX=hipcc cmake ${QUDA_PATH} -DQUDA_GPU_ARCH=sm_30 -DQUDA_DIRAC_STAGGERED=ON -DQUDA_DIRAC_DOMAIN_WALL=OFF -DQUDA_DIRAC_TWISTED_MASS=OFF -DQUDA_LINK_HISQ=OFF -DQUDA_MULTIGRID=ON -DQUDA_MPI=OFF -DQUDA_INTERFACE_MILC=OFF -DQUDA_QMP=ON -DQUDA_QMPHOME=${SCIDAC_PATH}/qmp -DQUDA_QIO=ON -DQUDA_QIOHOME=${SCIDAC_PATH}/qio -DQUDA_LIMEHOME=${SCIDAC_PATH}/qio -DQUDA_USE_EIGEN=ON -DQUDA_DOWNLOAD_EIGEN=OFF -DEIGEN_INCLUDE_DIR=${QUDA_PATH}/include/externals/eigen-hacked -DQUDA_BUILD_SHAREDLIB=ON -DQUDA_BUILD_ALL_TESTS=ON -DQUDA_TEX=OFF -DCMAKE_BUILD_TYPE=DEVEL -DQUDA_LIBS="-L/opt/hpc/software/mpi/hpcx/v2.4.1/gcc7.2.1/ompi/lib/ -lopen-pal /opt/rocm/lib/libhipblas.so"https://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/0ffc2c7c3e4e1921cbb3df8301a26f4be9552a35The changes based on hipify-perl-for-clang2019-11-03T14:58:36+08:00Yi-Boybyang@itp.ac.cnhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/3fa55816a49365941258f473f5379fe3bcd68f26Merge pull request #900 from lattice/feature/persistent_deflation_space2019-10-28T21:26:27-07:00maddyscientistmclark@nvidia.com
Feature/persistent deflation spacehttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/35d6f69e77c4e3c4451716c6886a127ed0fefb00ExposeSolver should return by reference and not by pointer2019-10-28T17:24:52-07:00maddyscientistmclark@nvidia.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/e13a6a7eae0efc0879eea02b91a7d7a28c2d94e1Improve run-time parameter checking when preserving deflation space, add more...2019-10-28T17:17:16-07:00maddyscientistmclark@nvidia.comImprove run-time parameter checking when preserving deflation space, add more detailed doxygen comments and make function name more QUDA-like
https://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/bbd2b254f601b0751896477d8bedc6686857cf9eDoxygen Markup2019-10-28T14:08:23-07:00cpviolatordmhowarth26@gmail.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/d8ce22aa073ec7e56fce0946b6c9456eff792f1eclang2019-10-28T14:05:17-07:00cpviolatordmhowarth26@gmail.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/90945e865f64ab2597b43c2749c62f146b147985removed old code2019-10-28T14:04:56-07:00cpviolatordmhowarth26@gmail.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/baee9dba968b53d86e3a017947017d59d1d73415Clang2019-10-28T13:50:53-07:00cpviolatordmhowarth26@gmail.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/a37b69c11071b5f93bc0ccf2b7669548cffd63d6Moved deflation objects to protected, neater calling sequence for MG deflation2019-10-28T13:50:12-07:00cpviolatordmhowarth26@gmail.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/d001eb48d1a0d10ab38cec33a194a75832808b5fDoxygen for deflation objects in Solver class2019-10-28T08:38:00-07:00cpviolatordmhowarth26@gmail.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/6a46102d6596e38d2469c5fb99f440796eaa04d4Doxygen for exposeSolver2019-10-28T08:26:37-07:00cpviolatordmhowarth26@gmail.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/4c146ebfa7a70445e5f5ec6efc25773a33bee31eRestore profile start/stop in TRLM constructor2019-10-28T00:15:50-07:00maddyscientistmclark@nvidia.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/56027fc354c5ef2ea8aa3a787cdbf5cbcf1c02d5Apply clang format2019-10-28T00:07:45-07:00maddyscientistmclark@nvidia.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/802120c2f5a46a401986d09630dc567abc52c3b7More removal of unnecessary copying and instead just move the pointers2019-10-28T00:05:11-07:00maddyscientistmclark@nvidia.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/5b29e545ed3eb906cc365d6debf1f42d9fdfb4a5Do not automatically disable null-space refreshment whenever we are preservin...2019-10-27T00:59:19-07:00maddyscientistmclark@nvidia.comDo not automatically disable null-space refreshment whenever we are preserving the deflation space - these are now decoupled options
https://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/0d8b338e1adbb28d9cb3fc5530d4dae788200bafFix ASAN warnings since CLI integration2019-10-26T23:58:19-07:00maddyscientistmclark@nvidia.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/39c4accbeb47d1a2fffd7b0cc62f7d1b77185454Fix buggy last commit2019-10-26T22:50:49-07:00maddyscientistmclark@nvidia.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/f06b9471c107eda29dd0115a8f1d356b8048684dFix ASAN error with GCR and CA-GCR deflated solvers2019-10-26T22:47:49-07:00maddyscientistmclark@nvidia.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/c7314f9a6ddc11833e42516dd5f630cdc3b5a33cremove integer comparison warning2019-10-26T17:07:15-07:00cpviolatordmhowarth26@gmail.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/4de269a014c8d0ea399bf28a942a372b0e85c455Merge branch 'feature/persistent_deflation_space' of...2019-10-26T17:04:32-07:00cpviolatordmhowarth26@gmail.comMerge branch 'feature/persistent_deflation_space' of <a href="https://github.com/lattice/quda" rel="nofollow noreferrer noopener" target="_blank">https://github.com/lattice/quda</a> into feature/persistent_deflation_space
https://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/fdb366f09a0965ed6a4697f0db9341c7a73bec13differentiate names in dump step2019-10-26T17:03:55-07:00cpviolatordmhowarth26@gmail.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/c322b54d98091bd89dcad3106ab2e272e9b2ccf9Apply clang format2019-10-26T00:46:53-07:00maddyscientistmclark@nvidia.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/e0efc49c68e44c956af15b31dbefe1217a2cda9fMinor cleanup of multigrid_evolve_test2019-10-26T00:16:37-07:00maddyscientistmclark@nvidia.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/c4a7b586aeff030c8d03c6228e0ee0886f20a186Verbosity tweaks2019-10-26T00:08:02-07:00maddyscientistmclark@nvidia.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/a1abc85fc71885e1f10b843ef77dbda19093b42fUpdate MG param structure when shifting mass/mu2019-10-25T22:30:37-07:00cpviolatordmhowarth26@gmail.comhttps://code.itp.ac.cn/ybyang/quda-with-hip/-/commit/1931ae707049a593ee61d66d70c3351586074c7bFix bug in mg CLI params2019-10-25T16:11:45-07:00maddyscientistmclark@nvidia.com