meson cannot find mkl_intel_thread #402
Description
Building with Intel oneAPI 2021.1 Classic Compilers and MKL currently fails with
meson/meson.build:77:4: ERROR: Fortran library 'mkl_intel_thread' not found
The meson log shows the following error:
Running compile:
Working directory: /home/awvwgk/projects/src/git/xtb/_build_oneapi/meson-private/tmpq76llaig
Command line: ifort /home/awvwgk/projects/src/git/xtb/_build_oneapi/meson-private/tmpq76llaig/testfile.f90 -o /home/awvwgk/projects/src/git/xtb/_build_oneapi/meson-private/tmpq76llaig/outpu
t.exe -D_FILE_OFFSET_BITS=64 -O0 -Wl,--start-group -lmkl_intel_thread -Wl,--end-group -Wl,--allow-shlib-undefined -diag-error 10006 -diag-error 10148 -diag-error 10155 -diag-error 10156 -dia
g-error 10157 -diag-error 10158 -diag-error 1292
Code:
stop; end program
Compiler stdout:
Compiler stderr:
ld: /opt/intel/oneapi/compiler/2021.1.1/linux/compiler/lib/intel64_lin/libifcoremt.a(for_vm.o): in function `for_allocate':
for_vm.c:(.text+0x426): undefined reference to `__kmpc_global_thread_num'
ld: for_vm.c:(.text+0x787): undefined reference to `__kmpc_global_thread_num'
ld: /opt/intel/oneapi/compiler/2021.1.1/linux/compiler/lib/intel64_lin/libifcoremt.a(for_vm.o): in function `for_alloc_allocatable':
for_vm.c:(.text+0xb03): undefined reference to `__kmpc_global_thread_num'
ld: for_vm.c:(.text+0xe5e): undefined reference to `__kmpc_global_thread_num'
ld: /opt/intel/oneapi/compiler/2021.1.1/linux/compiler/lib/intel64_lin/libifcoremt.a(for_vm.o): in function `for_deallocate':
for_vm.c:(.text+0x118d): undefined reference to `__kmpc_global_thread_num'
ld: /opt/intel/oneapi/compiler/2021.1.1/linux/compiler/lib/intel64_lin/libifcoremt.a(for_vm.o):for_vm.c:(.text+0x128d): more undefined references to `__kmpc_global_thread_num' follow
A workaround is to add -Dfortran_link_args=-qopenmp in the setup step, since the globally added link argument will bleed into the command options used to find the libraries in meson. Not the best solution though.
Opening this issue to document the current situation and possible workarounds. Thanks to @cbannwarth for bringing this issue to my attention.