WIP: Update ARM Toolchain and libopencm

[rpls]

• PR changes testvectors
• Tests pass in qemu
• Testvectors pass in qemu
• Tests pass on Nucleo-L4R5ZI
• Testvectors pass on Nucleo-L4R5ZI
• Updated Benchmarks
• Updated Skiplist entries

[rpls]

Well, sphincs doesn't build anymore, and falcon-1024 fails its testvectors on QEMU...

[mkannwischer]

Wow. build failures can happen, but testvector failures are rather worrying.

[rpls]

So for sphincs we need a fix in PQClean/PQClean#582, for falcon-{512,1024}-clean we'll need to investigate. The m4,opt-ct, ..., variants work fine.

[mkannwischer]

@rpls - the fix for SPHINCS+ has been merged upstream in PQClean/PQClean#584.
I updated to that commit just now.

Could you do another rounds of tests and benchmarks, please?
There is some risk that benchmarks fail for some schemes due to #376.

[rpls]

So far, a couple of weird failures. Some of the mirith variants hang indefinitely. mayo{1,3}-m4f speed tests fail??

[mkannwischer]

I know why MAYO fails: this version of the code does not handle sm=m correctly. We have fixed that in round 2. (Note that the speed.c was changed only recently to include checking of the verification result).
May be a similar issue for the other weird failure.

I will update MAYO soon™️.

Can you figure out if the other failures are also caused by this (simply using two buffers should be sufficient)? If so, we should inform the submitters to fix their code.

[mkannwischer]

I know why MAYO fails: this version of the code does not handle sm=m correctly. We have fixed that in round 2. (Note that the speed.c was changed only recently to include checking of the verification result). May be a similar issue for the other weird failure.

I will update MAYO soon™️.

@rpls - MAYO is updated in #391. So that should be unblocked now.

[rpls]

I did a rebase to include all of the previous updates, and I'll run the benchmarks over the weekend. I have three boards, so I can split it up to some degree.

[mkannwischer]

I did a rebase to include all of the previous updates, and I'll run the benchmarks over the weekend. I have three boards, so I can split it up to some degree.

I have not touched anything in mirith - that should be checked first.

[rpls]

diff.ods

[mkannwischer]

diff.ods

How many iterations is this? 68% worse performance is hard to believe.