[LV] is it appropriate to add comment to the final output assembly code indicating the VF value selected ？

Beginners
1

In the vectorization phase, we often select specific VF by instruction cost, so it’s not obvious which VF we choose in the final assembly code, especially in the case of template functions for SVE (because they unified use register z registers in AArch64), and the source code line numbers are also duplicated in different versions of specialization function.
For example, the following 2 functions come from the same template function eval, have mixed types float and double in the kernel loop body, it is difficult to infer the current VF value directly from assembly code without carefully examining the functionality of the current assembly.

eval<0,0,1,float,double>(...)
eval<0,1,1,float,double>(...)

So is it appropriate to add comment to the final output assembly code indicating the time VF value selected for the current loop?

2

Not really a direct answer to your question, but maybe nice to know: You can look at the “optimization record” to see what loop was vectorized with which vectorization factor: clang -O3 -fsave-optimization-record foo.c will create a foo.opt.yaml file in which you can search for “loop-vectorize” entries until you find the one for the loop you want to check and then you will see " - VectorizationFactor: …" somewhere.

3

You can also try:

> clang -O3 -Rpass=loop-vectorize loops.c

Then you only remarks from successful vectorisations. There are variants with missed. Getting remarks/metadata into assembler is a different story.