icpx: warning: overriding '-ffp-contract=fast' option with '-ffp-contract=on' [-Woverriding-t-option]

[valassi]

This is for clang14 (also 13?), or for icpx that uses clang14.

I think that it is because I am partly disabling fast math on one file for c++, but cuda always keeps it

ccache /usr/local/cuda-11.7/bin/nvcc  -O3  -lineinfo -I. -I../../src -I../../../../../tools -I/usr/local/cuda-11.7/include/ -DUSE_NVTX -gencode arch=compute_70,code=compute_70 -gencode arch=compute_70,code=sm_70 -use_fast_math -std=c++17  -ccbin /cvmfs/projects.cern.ch/intelsw/oneAPI/linux/x86_64/2022/compiler/2022.1.0/linux/bin/icpx -allow-unsupported-compiler -DMGONGPU_FPTYPE_DOUBLE -Xcompiler -fno-fast-math -Xcompiler -fPIC -c gCrossSectionKernels.cu -o gCrossSectionKernels.o
clang++: warning: overriding '-ffp-contract=fast' option with '-ffp-contract=on' [-Woverriding-t-option]
clang++: warning: overriding '-ffp-contract=fast' option with '-ffp-contract=on' [-Woverriding-t-option]
clang++: warning: overriding '-ffp-contract=fast' option with '-ffp-contract=on' [-Woverriding-t-option]

This is llvm/llvm-project issues/54625

[valassi]

I added a patch to avoid conflicting fast math and no fast math, but this is not enough: even just specifying no fast math triggers the warning?!

icpx  -O3  -std=c++17 -I. -I../../src -I../../../../../tools -DUSE_NVTX -Wall -Wshadow -Wextra   -march=skylake-avx512 -mprefer-vector-width=256  -DMGONGPU_FPTYPE_DOUBLE -fno-fast-math -I/usr/local/cuda-11.7/include/ -fPIC -c CrossSectionKernels.cc -o CrossSectionKernels.o
clang++: warning: overriding '-ffp-contract=fast' option with '-ffp-contract=on' [-Woverriding-t-option]

I therefore added instead a -Wno-overriding-t-option... I am not sure it is a good idea (maybe I should leave the warning?) but I will go for this for the moment? Or I will keep it there disabled for now.

[valassi]

I think this has disappeared from clang14 builds on alma9 (I might be wrong).

But I confirm it is still there for icx2023 builds on alma9

icpx: warning: overriding '-ffp-contract=fast' option with '-ffp-contract=on' [-Woverriding-t-option]

Keep this open and wait for a real fix rather than disabling the warning

[valassi]

This is still there in clang14 on alma9

ccache /data/avalassi/GPU2023/madgraph4gpuX/tools/compilers/mg-lcg-clang++-14.0.6  -O3  -std=c++17 -I. -I../../src -Wall -Wshadow -Wextra -ffast-math  -fopenmp -march=haswell  -DMGONGPU_FPTYPE_DOUBLE -DMGONGPU_FPTYPE2_DOUBLE -fno-fast-math -fno-fast-math -fPIC -c CrossSectionKernels.cc -o CrossSectionKernels.o
clang-14: warning: overriding '-ffp-contract=fast' option with '-ffp-contract=on' [-Woverriding-t-option]

See issue 54625 in https://github.com/llvm/llvm-project/issues. This correctly points to the fact that this is a "the last flag wins" situation. I am passing two flags, with and ithout fast math. I should remove one of them, it's as easy as that.