Compiler feature selection broke AWFY Cranelift testing
Categories
(Core :: JavaScript: WebAssembly, defect, P2)
Tracking
()
Tracking | Status | |
---|---|---|
firefox-esr68 | --- | unaffected |
firefox74 | --- | unaffected |
firefox75 | --- | unaffected |
firefox76 | --- | fixed |
People
(Reporter: bbouvier, Assigned: bbouvier)
References
(Regression)
Details
(Keywords: regression)
Attachments
(1 file)
See also https://arewefastyet.com/linux64/wasm-misc?numDays=60. Cranelift has had the same performance as Ion for a few days now, and since Cranelift hasn't been updated recently, I strongly suspect the compiler feature selection work caused this.
Once features can all be disabled through the command-line and through about:config flags, then we should be able to change the configuration of the benchmarking system so that it runs with the right preferences, and things should get back to normal. So I think this is blocked by bug 1620986, and then the next item would be making sure there are browser about:config prefs to disable the same features too.
Comment 1•4 years ago
|
||
Yeah, sorry about that. I'm waiting for wingo to re-re-land the patch that allows me to create a sensible switch without forcing him to rebase painfully and me to write a bunch of code that will be thrown away immediately.
As I said in email the other day, for now cranelift cannot be tested unless multi-value is disabled at compile time, and shared memory must be disabled at compile time or run time.
Updated•4 years ago
|
Comment 2•4 years ago
|
||
I think this was not actually blocked on bug 1620986, since wasm multi-value is currently only enabled in the shell. It ought to be enough to disable shared memory in Nightly to test cranelift. A quick test with wasm_verbose=true, wasm_cranelift=true, wasm_baseline=false, and shared_memory=false indicates that that is indeed the case: the spew in the console shows cranelift being available and wasm compilation appears to succeed on a simple test case.
Assignee | ||
Comment 3•4 years ago
|
||
Nice, it actually simplifies the problem a bit.
Assignee | ||
Comment 4•4 years ago
|
||
Updated•4 years ago
|
Pushed by bbouvier@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/7be47bd25f46 disable Ion and shared memory when benchmarking Cranelift; r=perftest-reviewers,sparky
Comment 6•4 years ago
|
||
bugherder |
Updated•4 years ago
|
Updated•4 years ago
|
Description
•