Jorgen's addition of HIP/CUDA abstraction to codegen and all processes (with extra fixes and tests: "jt774" replacing PR #774)#801
Merged
valassi merged 98 commits intomadgraph5:masterfrom valassi:jt774Jan 31, 2024
+21,530-16,575
Commits
Commits on Aug 24, 2023
- committed
- committed
- committed
- committed
- committed
- committed
- committed
Commits on Sep 28, 2023
Commits on Sep 29, 2023
Commits on Jan 24, 2024
- committed
- committed
- committed
- committed
- committed
- committed
- committed
- committed
- committed
- committed
Commits on Jan 25, 2024
[jt774] cherry-pick commit 1b5c0fd ([jthip] backport to CODEGEN from ggttgg.mad on Tue Jul 18 18:11:04 2023 +0200)
committed- committed
- committed
- committed
- committed
- committed
- committed
- committed
- committed
- committed
- committed
- committed
- committed
- committed
- committed
- committed
Commits on Jan 26, 2024
Commits on Jan 27, 2024
- committed
- committed
- committed
[jt774] in gg_tt.sa, bypass std::filesystem completely to ease portability on LUMI madgraph5#803 (and undo previous changes to address this issue in __HIPCC__)
committed[jt774] first tput tee test on LUMI with AMD GPU (using ggtt.sa): here fgcheck.exe fails because of link issues madgraph5#802
committed[jt774] in gg_tt.sa cudacpp.mk, use FC to link fgcheck.exe, gcheck.exe, runTest.exe when hipcc is used madgraph5#802
committed- committed
[jt774] second (first successful) tput tee test on LUMI with AMD GPU (using ggtt.sa), after fixing fgcheck.exe link issues madgraph5#802
committed- committed
- committed
- committed
[jt774] in gg_tt.mad cudacpp.mk, add -lpthread when linking runTest.exe using $FC=gfortran on LUMI/HIP madgraph5#802
committed[jt774] first (successful) tput tee test on LUMI with AMD GPU using ggtt.mad, after adding pthread in the link madgraph5#802
committed[jt774] first tmad test on LUMI - it fails for Fortran vs C++ (no HIP)... why? due to link issues madgraph5#802 again?
committed- committed
- committed
- committed
[jt774] in CODEGEN cudacpp.mk, backport form gg_tt.mad, add -lpthread when linking runTest.exe using $FC=gfortran on LUMI/HIP madgraph5#802
committed- committed
- committed
- committed
- committed
- committed
- committed
- committed
Commits on Jan 28, 2024
Commits on Jan 29, 2024
[jt774] in gq_ttq.mad cudacpp.mk, allow multi-word CXX (e.g. for the Cray compiler on LUMI madgraph5#807) in HIP builds.
committed[jt774] in CODEGEN (backport gq_ttq.mad) cudacpp.mk, allow multi-word CXX (e.g. for the Cray compiler on LUMI madgraph5#807) in HIP builds.
committed- committed
- committed
- committed
- committed
[jt774] manually fix all processes for cudacpp '-G' option in HIP madgraph5#808 (and also for Jorgen's name in copyright)
committed[jt774] in gg_tt.mad, fix mgOnGpuCxtypes.h so that cucomplex and thrust are only used for cuda and not for hip (madgraph5#810)
committed[jt774] in gg_tt.mad, fix complex type #ifdef's and fix "-x hip" compiler options for HIP (madgraph5#810)
committed
Commits on Jan 30, 2024
[jt774] in CODEGEN (backport from gg_tt.mad), fix mgOnGpuCxtypes.h so that cucomplex and thrust are only used for cuda and not for hip (madgraph5#810) and fix complex type #ifdef's and fix "-x hip"…
committed- committed
- committed
- committed
- committed
- committed
- committed
Commits on Jan 31, 2024
- committed
- committed
- committed
- committed
- committed
- committed
- committed
