1359820 - wasm: arm-vfp testing is completely broken

Status: RESOLVED WORKSFORME

(Core :: JavaScript: WebAssembly, defect, P5)

Description

[Benjamin Bouvier [:bbouvier] (inactive)]

STR: try to run the jit-test suite with the --args=--arm-hwcap=vfp on the wasm tests.

There are many failures at the moment. They probably reduce to a handful of causes, but it'd be nice to fix them at some point, to make arm-vfp fixing easier in the future.

Really just a nice-to-have because this is not a tier-1 platform, but this might hide bugs related to idiv for instance, which is still common.

Comment 1

[Lars T Hansen [:lth]]

I understand the desire to test vfp-only (as opposed to vfpv3, etc), but IIUC this is an ARMv6 configuration.  Wouldn't it be better to include additional flags so that we're at least testing ARMv7?

Comment 2

[Benjamin Bouvier [:bbouvier] (inactive)]

(In reply to Lars T Hansen [:lth] from comment #1)
> I understand the desire to test vfp-only (as opposed to vfpv3, etc), but
> IIUC this is an ARMv6 configuration.  Wouldn't it be better to include
> additional flags so that we're at least testing ARMv7?

Yes. I think the actual interesting thing to test is an ARM configuration where we don't have idiv.

What kind of ARM hardware capabilities string are people using, in general?

Comment 3

[Lars T Hansen [:lth]]

This is the string I've been using as my default:

  ARMHWCAP=armv7,vfp,vfpv3,vfpv4,neon,idiva,hardfp

I worked this out at a time when the default HW cap string for the simulator was something fairly feeble without idiv, etc, and my belief at the time was that it represented a "current" fairly high-end ARMv7 device.  I think the default has since changed to be closer to what's above.

Comment 4

[BMO Automation]

Per policy at https://wiki.mozilla.org/Bug_Triage/Projects/Bug_Handling/Bug_Husbandry#Inactive_Bugs. If this bug is not an enhancement request or a bug not present in a supported release of Firefox, then it may be reopened.

Comment 5

[Lars T Hansen [:lth]]

The default in the simulator is now something sane (armv7 + lots of FP flags).

There's still an issue that the env var / cmdline switch syntax does not allow for defaults-plus-something or defaults-minus-something (I'm thinking ARMHWCAP='default-neon+hardfp' would be a nice syntax) but then we should open a new issue for that in the JS:JIT component.