Closed valassi closed 6 months ago
Removing -G in this single file an dbuilding by hand is enough
[valassia@nid005377 bash] ~/GPU2023/madgraph4gpuX/epochX/cudacpp/gq_ttq.mad/SubProcesses/P1_gu_ttxu > ccache /opt/rocm/bin/hipcc -g -O0 -I. -I../../src -I/opt/rocm/include/ -target x86_64-linux-gnu --offload-arch=gfx90a -DHIP_FAST_MATH -DHIP_PLATFORM=amd -fPIC -std=c++17 -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE -fPIC -c gRamboSamplingKernels.cu -o gRamboSamplingKernels.o
Note: the problem is that CUOPTFLAG really refers to CUDA, but is also applied to HIP, it must be removed.
debug: OPTFLAGS = -g -O0
debug: CUOPTFLAGS = -G
debug: MAKEDEBUG := debug
debug: all.$(TAG)
This is also fixed in PR #801 https://github.com/madgraph5/madgraph4gpu/pull/801/commits/6572c5d0afe9c26675e78876c4e68cab5206b49e
Still related to PR #801 HIP support, and to debugging the crash #806 n gqttq.
The 'make -f cudacpp.mk debug' fails for HIP
(Note: the readFromKFD error is always there and is harmless - I see it in every single LUMI build, all over the place)