ulan opened issue #4010:
After upgrading to version 0.35.2 we are seeing register allocation panics in our integration tests:
thread '<unnamed>' panicked at 'register allocation: Analysis(EntryLiveinValues([v430V]))', cranelift-codegen-0.82.2/src/machinst/compile.rs:98:10
The panic seems to be fixed on the
main
branch.The regression bisects to https://github.com/bytecodealliance/wasmtime/pull/3682 and the fix bisects to https://github.com/bytecodealliance/wasmtime/pull/3934, which looks related because
v430V
appears inucomiss
instructions:Inst 893: cmovnzq %v588J, %v429J; cmovpq %v588J, %v429J Inst 894: xorps %v585V, %v585V Inst 895: ucomiss %v585V, %v430V Inst 896: setp %v586Jb
I can share the complete output of
VCode_ShowWithRRU
if that would help.I don't have a standalone Wasm file that reproduces this panic, but I can try to extract it from the integration test if that would help.
ulan labeled issue #4010:
After upgrading to version 0.35.2 we are seeing register allocation panics in our integration tests:
thread '<unnamed>' panicked at 'register allocation: Analysis(EntryLiveinValues([v430V]))', cranelift-codegen-0.82.2/src/machinst/compile.rs:98:10
The panic seems to be fixed on the
main
branch.The regression bisects to https://github.com/bytecodealliance/wasmtime/pull/3682 and the fix bisects to https://github.com/bytecodealliance/wasmtime/pull/3934, which looks related because
v430V
appears inucomiss
instructions:Inst 893: cmovnzq %v588J, %v429J; cmovpq %v588J, %v429J Inst 894: xorps %v585V, %v585V Inst 895: ucomiss %v585V, %v430V Inst 896: setp %v586Jb
I can share the complete output of
VCode_ShowWithRRU
if that would help.I don't have a standalone Wasm file that reproduces this panic, but I can try to extract it from the integration test if that would help.
ulan commented on issue #4010:
@abrown @cfallin: Should https://github.com/bytecodealliance/wasmtime/pull/3934 be cherry-picked to v0.35?
cfallin commented on issue #4010:
@ulan thanks for this report; it certainly looks like the bug that we fixed in #3934. I'm trying to work out now what happened, and it looks like at the time we made the fix, we missed that the original bug was old enough to have gone out in the 0.35 release. I will go ahead with the patch-release process and prepare 0.35.3 with the fix.
ulan commented on issue #4010:
Thanks for the fast response and cherry-pick @cfallin!
cfallin commented on issue #4010:
@ulan this workflow is running now to publish new crates for 0.35.3; please let us know if there are any other issues! I'll go ahead and close this issue now.
cfallin closed issue #4010:
After upgrading to version 0.35.2 we are seeing register allocation panics in our integration tests:
thread '<unnamed>' panicked at 'register allocation: Analysis(EntryLiveinValues([v430V]))', cranelift-codegen-0.82.2/src/machinst/compile.rs:98:10
The panic seems to be fixed on the
main
branch.The regression bisects to https://github.com/bytecodealliance/wasmtime/pull/3682 and the fix bisects to https://github.com/bytecodealliance/wasmtime/pull/3934, which looks related because
v430V
appears inucomiss
instructions:Inst 893: cmovnzq %v588J, %v429J; cmovpq %v588J, %v429J Inst 894: xorps %v585V, %v585V Inst 895: ucomiss %v585V, %v430V Inst 896: setp %v586Jb
I can share the complete output of
VCode_ShowWithRRU
if that would help.I don't have a standalone Wasm file that reproduces this panic, but I can try to extract it from the integration test if that would help.
Last updated: Oct 23 2024 at 20:03 UTC