GPU checkpoint compilation failed.

[GYDmedwin]

Hi,
I tried to compile the GPU checkpoint module using make install-gpu.

However, the compilation always fails. The problem I'm facing now is a library link problem：

g++ -std=gnu++11 -fpermissive -g -lncurses -lpthread -lm -lz -ldl -lexpat -llzma -lmpfr -Wl,--dynamic-list=utils/proc-service.list -o cricket build/cricket-cr.o build/cricket-device.o build/cricket-elf.o build/cricket-file.o build/cricket-heap.o build/cricket-register.o build/cricket-stack.o build/cricket-util.o build/gdb.o build/log.o build/main.o ../submodules/lib/libgdb.a ../submodules/lib/libbfd.a ../submodules/lib/libiberty.a ../submodules/lib/libreadline.a ../submodules/lib/libdecnumber.a ../submodules/lib/libcudacore.a ../submodules/lib/libopcodes.a ../submodules/lib/libgnu.a
/usr/bin/ld: ../submodules/lib/libgdb.a(tui-win.o): warning: relocation against curscr' in read-only section .text'
/usr/bin/ld: ../submodules/lib/libgdb.a(target-float.o): in function mpfr_float_ops::gdb_mpfr::gdb_mpfr(mpfr_float_ops::gdb_mpfr const&)': /data/guyida/cricket/submodules/cuda-gdb/build/gdb/../../gdb/target-float.c:1213: undefined reference to mpfr_init2'
/usr/bin/ld: /data/guyida/cricket/submodules/cuda-gdb/build/gdb/../../gdb/target-float.c:1213: undefined reference to mpfr_init2' /usr/bin/ld: ../submodules/lib/libgdb.a(target-float.o): in function mpfr_float_ops::from_target(floatformat const*, unsigned char const*, mpfr_float_ops::gdb_mpfr&) const':
/data/guyida/cricket/submodules/cuda-gdb/build/gdb/../../gdb/target-float.c:1284: undefined reference to mpfr_add' /usr/bin/ld: ../submodules/lib/libgdb.a(target-float.o): in function mpfr_float_ops::gdb_mpfr::~gdb_mpfr()':
/data/guyida/cricket/submodules/cuda-gdb/build/gdb/../../gdb/target-float.c:1218: undefined reference to mpfr_clear' /usr/bin/ld: /data/guyida/cricket/submodules/cuda-gdb/build/gdb/../../gdb/target-float.c:1218: undefined reference to mpfr_clear'
/usr/bin/ld: ../submodules/lib/libgdb.a(target-float.o): in function mpfr_float_ops::from_target(floatformat const*, unsigned char const*, mpfr_float_ops::gdb_mpfr&) const': /data/guyida/cricket/submodules/cuda-gdb/build/gdb/../../gdb/target-float.c:1296: undefined reference to mpfr_set_zero'
/usr/bin/ld: ../submodules/lib/libgdb.a(target-float.o): in function mpfr_float_ops::gdb_mpfr::gdb_mpfr(mpfr_float_ops::gdb_mpfr const&)': /data/guyida/cricket/submodules/cuda-gdb/build/gdb/../../gdb/target-float.c:1213: undefined reference to mpfr_init2'
/usr/bin/ld: ../submodules/lib/libgdb.a(target-float.o): in function `mpfr_float_ops::from_target(floatformat const*, unsigned char const*, mpfr_float_ops::gdb_mpfr&) const':
... ...

I checked the libraries in /usr/lib/x86_64 and found that all the required libraries are there, so I guess it may be a library version problem.

Could you please provide the versions of the dependent libraries you compiled with? Or is there any other way (Docker?) to test the checkpoint/restore function?

THAKS！

[n-eiling]

You do not need to compile the gpu part. Checkpoint/Restore between CUDA kernel calls is implemented in the cpu folder. gpu contains the in-kernel checkpoint/restore functionality. This works only for Pascal generation GPUs, so is quite outdated.