Disable BB GMP and MPFR, to fix performance regressions
#32102
Conversation
|
@nanosoldier Edit: |
|
RIP me. |
|
@nanosoldier |
|
Long live Nanosoldier! |
|
Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan |
|
Why are you guys just testing LinAlg? @nanosoldier |
|
Also we are not using BB for the release builds are we? |
|
I dunno, but it is important that what ends up in the binary is built the same way that what we test performance on. |
We are using BB for the official releases. No reason not to, as far as we can tell. |
|
IMO we should rework nanosoldier to use the binaries built by the buildbots. |
Hah yeah, I did it because it's what Fredrik did and I'm a 🐑. He apparently noticed that linalg wasn't the best choice given the edit. |
|
I think the only dependency that Nanosoldier doesn't use BB for is LLVM. |
|
Well, I copied you without looking carefully: #32097 (comment) :) |
|
We're all 🐑 |
|
Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan |
|
The relevant stuff looks good. |
|
The memory regressions are weird... But they should not be related. |
|
So should we just merge this into 1.3 + master as well since it seems like it is here to stay for a while? |
|
Unless we can fix the issue somehow. @staticfloat, any thoughts? |
|
How about we mark it for triage? So that if we hit a couple of RCs and not resolved, we can at least merge into 1.3 then. |
|
I put #31759 on the 1.3 milestone at least. |
Let's see if this compiles everywhere (fingers crossed for FreeBSD) and gives us the bigfloat performance back.