Reproductibility

[tperchoc]

Hello,
Thank you for your work, just
I’m having difficulties replicating the example FIR kernel mapping from fir.cpp to kernel.bc and achieving the expected simple DFG with II=1.
Despite disabling optimizations, extracting only the kernel function, and adjusting param.json, I always get a more complex DFG and higher II.
Could you please clarify the exact steps, compiler commands, and LLVM version you use to produce the kernel.bc shown in your documentation?
I
Thank you

What i get :

What i should have :

I believe its comming from fir.cpp -> kernel.bc, i tried :
clang-12 -O0 -emit-llvm -c fir.cpp -o kernel.bc
clang-12 -O0 -fno-inline -fno-unroll-loops -fno-vectorize -fno-slp-vectorize
-emit-llvm -c fir.cpp -o kernel.bc
clang-12 -O0 -x c -std=c99 -fno-builtin -fno-inline -fno-unroll-loops
-fno-vectorize -fno-slp-vectorize -emit-llvm -c fir.cpp -o kernel.bc

[tancheng]

Plz try -O3 for clang.

@MeowMJ plz help if O3 doesn't work. Thanks~!

[tperchoc]

Thank you for responding so quickl
Okay with O3 i got this do you think its good ?

[tancheng]

The DFG is from unrolling the loop, we need to disable the unrolling. Did you use clang-12 -O3 -fno-inline -fno-unroll-loops -fno-vectorize -fno-slp-vectorize -emit-llvm -c fir.cpp -o kernel.bc?

[tperchoc]

I just used the command I found while searching in CGRA-Flow and it works, thank you very much