Get rid of ykconfig -ao.
#1411
Get rid of ykconfig -ao.
#1411
Conversation
bin/yk-config
Outdated
| debug*) OUTPUT="$OUTPUT -g" ;; | ||
| esac | ||
|
|
||
| # Disable backend optimisations that Yk can't handle by adding the |
There was a problem hiding this comment.
Just noticed a mistake here. We have a linker flag in the --cflags section!
Will fix and retest.
tests/c/aot_debuginfo.c
Outdated
| @@ -1,4 +1,5 @@ | |||
| // ignore-if: test $YK_CARGO_PROFILE != "debug" | |||
| // Compiler: | |||
| // env-var: YKT_EXTRA_CC_FLAGS=-O1 -g | |||
There was a problem hiding this comment.
I think this should be YKB_EXTRA_...?
There was a problem hiding this comment.
YKT was traditionally used for testing only, but I see it's been renamed/removed at some point.
You can't use this variable outside of tests since it requires running the compiler via the compiler wrapper. Is that OK still?
There was a problem hiding this comment.
Should we just rename it to YKB since it's a build-time option (even if only "build testing")? If you think that's the way to go, you can rename + squash in one go.
There was a problem hiding this comment.
OK.
I must admit I get confused at times with our variables. I've seen so many reincarnations of them that I'm just constantly lost!
There was a problem hiding this comment.
I know the feeling! FWIW, I think the prefixes are now documented in the book tolerable well?
There was a problem hiding this comment.
Yeah mostly. There's a couple of things that often still catch me out:
- variables with no
DorB, likeYK_LOG. I always typeYKD_LOG. - the mandatory
<path>:prefix ofYKD_LOG_IR, because I was spoiled by the optional prefix ofYK_LOG:)
There was a problem hiding this comment.
the mandatory : prefix of YKD_LOG_IR, because I was spoiled by the optional prefix of YK_LOG :)
Hmm, you're right, that shouldn't be mandatory! PR welcome!
There was a problem hiding this comment.
I know. It's just such a small fish right now!
There was a problem hiding this comment.
Anyway YKT -> YKB in a04c8f8
|
Other than |
|
Please squash. |
|
I've also sneaked in 3eda80d which will be needed for the yklua tests when we re-enable them. Its a flag we already have enabled for link time, but is also needed for pre-link time, since that pas can (and does) run at both times. |
|
Yes I saw it. Please squash. |
|
squashed. |
github-merge-queue
bot
removed this pull request from the merge queue due to failed status checks
|
My apologies. I force pushed a change to sync in ykllvm too. Ready. |
github-merge-queue
bot
removed this pull request from the merge queue due to failed status checks
|
Just a test failing due to SWT recorder calls. Fixed in 7187176 OK to squash? |
|
Please squash. |
Not so long ago we added `yk-config -ao <level>` to allow us to run our
pre-canned franken-O1 pipeline for the entire test suite. This was only
ever designed as an interim hack, and we've wanted Yk to simply work
with standard optimisation flags like -O1, -O2 etc.
This change is a step in that direction.
We remove `-ao` and our non-standard defaults and add the ability for
individual tests to specify extra compiler flags. This gives us the
ability to specify optimisation parameters on a per-test basis.
I then ported as many tests as were practical to use -O1. The remaining
tests (that were either too involved for me to convert right now, or
actually rely on an IR shape generated by our old default pipeline) I've
left using our franken-O1 by passing:
-O0 -Xclang -disable-O0-optnone -Xlinker \
--lto-newpm-passes=instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>
We do have to disable yklua in CI for now, as yklua still uses
`yk-config -ao` and we have to break a circular dependency.
|
squashed |
Requires: ykjit/ykllvm#209 (so will need a push to re-sync ykllvm)
Not so long ago we added
yk-config -ao <level>to allow us to run our pre-canned franken-O1 pipeline for the entire test suite. This was only ever designed as an interim hack, and we've wanted Yk to simply work with standard optimisation flags like -O1, -O2 etc.This change is a step in that direction.
We remove
-aoand our non-standard defaults and add the ability for individual tests to specify extra compiler flags. This gives us the ability to specify optimisation parameters on a per-test basis.I then ported as many tests as were practical to use -O1. The remaining tests (that were either too involved for me to convert right now, or actually rely on an IR shape generated by our old default pipeline) I've left using our franken-O1 by passing:
-O0 -Xclang -disable-O0-optnone -Xlinker
--lto-newpm-passes=instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>
We do have to disable yklua in CI for now, as yklua still uses
yk-config -aoand we have to break a circular dependency.