when I look vectorize_loop.cc I found that it is all about loop operations.After we specify the x86 backend, what does tvm do to generate SIMD instructions?Where is the specific code?
TVM generates a LLVM JIT module first. The LLVM module will be lowered to X86 target when necessary.
If you use
module.save("xxx.o") to dump it to an object file, and your native machine is X86,
you can use
gcc-objdump -d xxx.o to see the generated X86.