Re: Are big tests allowed in binutils?

[Jan Beulich <jbeulich@suse.com>]

On 21.06.2022 18:28, Dmitry Selyutin via Binutils wrote:
> The tests there are already generated with this assumption in mind
> already, they don't check all possible combinations (otherwise we'd
> have ended up with ~4GiB per test, eh).
> At the same time, they also check the operands together, basically
> this is a bunch of nested for loops, where each loop checks several
> predefined values for each operand.

This (in particular "nested loops") doesn't sound like it's matching
what Richard has suggested (which is a pattern I also generally try
to follow). Taking his example and assuming that I properly interpret
"several predefined values", your approach would result in

	FOO r0, r0
	FOO r0, r15
	FOO r15, r0
	FOO r15, r15

i.e. one more test than necessary. Obviously the number of "excess"
tests would grow (significantly) with the number of operands an insn
has (while there would be no difference for single-operand insns).

Jan

> I want to write some code which generates a test generator for
> binutils; this is the simplest option to avoid binutils depending on
> our code (you for sure don't want it).
> 
> This, however, is a big task, it requires many changes on our side,
> and I don't have time for this right now. It's also blocked somewhat
> with another task.
> Good news is that I will undoubtedly do it, because we have a plethora
> of scenarios which depend on auto-generation.
> And yes, we do want to be good citizens, that's why I brought up the
> whole auto-generation topic. :-)
>