config/arm: use eabihf for hard float #4023
Conversation
config/arch.aarch64
Outdated
| @@ -13,7 +13,7 @@ | |||
| generic|cortex-a35|cortex-a53|cortex-a57|cortex-a72|exynos-m1|qdf24xx|thunderx|xgene1|cortex-a57.cortex-a53|cortex-a72.cortex-a53) | |||
| TARGET_SUBARCH=aarch64 | |||
| TARGET_VARIANT=armv8-a | |||
| TARGET_ABI=eabi | |||
| TARGET_ABI=eabihf | |||
There was a problem hiding this comment.
huh? there is no soft float abi on aarch64.
There was a problem hiding this comment.
Sooo... are you agreeing with this change? Regardless of whether there is a soft float abi or not, the suggestion is that we should be including the hf suffix in the hard float abi.
There was a problem hiding this comment.
The aarch64 change is probably unnecessary. zlib-ng's hardfloat test appears to be looking for eabihf only for arm. There's a check further down which assumes floating point is true on aarch64. No floating point support (march=armv8-a+nofp(+nosimd?)) is possible, but I haven't seen it in the wild.
There was a problem hiding this comment.
It looks required to me - this is the zlib-ng code in question:
where BASEARCH_ARM_FOUND is:
Therefore setting hard or soft float by matching eabihf occurs for BOTH arm AND aarch64.
I can test this later today and confirm what happens (I may have already done this, but forgotten!) - I can build Allwinner/A64 (which is using aarch64) with eabi (ie. without this PR) and I'm fairly certain zlib-ng will try to use soft float. Allwinner/A64 definitely builds successfully with this PR and zlib-ng.
There was a problem hiding this comment.
By the way I'm not saying what zlib-ng is doing is entirely correct or sensible (it's certainly wrong in places with bias towards armv8 etc.), but if hard float should be reflected in the os triplet then that's something we're not currently doing and can add easily enough hopefully without any issues.
There was a problem hiding this comment.
arm change is ok. aarch64 is not and should not be added. it makes no sense to add just because random library's cmake buildsystem is (semi-)broken or makes wrong assumptions.
There was a problem hiding this comment.
Yes, we've been discussing this on LE Slack and have come to the same conclusion, I'll remove the aarch64 change. BTW I've posted our/my concerns to zlib-ng, here - would appreciate your thoughts if want to post there! 😄
There was a problem hiding this comment.
aarch64 change now dropped.
There was a problem hiding this comment.
well, they could just trust distro's compiler / CFLAGS.
There was a problem hiding this comment.
Yes, that's what I suggested. 😄
|
I've been testing this for a few days now for a couple of builds on Allwinner/H6. Looks good to me, no regressions here |
Following a discussion on zlib-ng/zlib-ng#478 it became apparent that the TARGET_ABI used for arm doesn't reflect the presence of hard float.
I have successfully created images for RPi, RPi2 and Allwinner/A64 with this PR (RPi and RPi2 images also booted OK). Add-ons are yet to be build tested.
Creating as a draft for discussion, and in case any additions are required.