abrown opened Issue #2432:
@jlb6740, I am opening to track a failure I noticed with x64 SIMD: https://github.com/bytecodealliance/wasmtime/pull/2428/checks?check_run_id=1421136203#step:7:1582
abrown commented on Issue #2432:
Looks like the issue is related to
f64x2.abs
:(assert_return (invoke "f64x2.abs" (v128.const f64x2 0x1p-1074 0x1p-1074)) (v128.const f64x2 0x0.0000000000001p-1022 0x0.0000000000001p-1022))
abrown edited Issue #2432:
@jlb6740, I am opening this issue to track a failure I noticed with x64 SIMD: https://github.com/bytecodealliance/wasmtime/pull/2428/checks?check_run_id=1421136203#step:7:1582
abrown commented on Issue #2432:
Looks like another instance of this: https://github.com/bytecodealliance/wasmtime/runs/1476300496?check_suite_focus=true#step:7:1883.
abrown commented on Issue #2432:
And another one reported by @alexcrichton in #2470 with
splat
: https://github.com/bytecodealliance/wasmtime/runs/1489518375?check_suite_focus=true#step:7:1807
abrown edited a comment on Issue #2432:
Looks like another instance of this: https://github.com/bytecodealliance/wasmtime/runs/1476300496?check_suite_focus=true#step:7:1883. It has multiple failures:
failures: wast::Cranelift::spec::simd::simd_f32x4 wast::Cranelift::spec::simd::simd_f64x2 wast::Cranelift::spec::simd::simd_f64x2_arith wast::Cranelift::spec::simd::simd_load wast::Cranelift::spec::simd::simd_splat
jlb6740 commented on Issue #2432:
@abrown Hi .. Thanks. It's hard to say if this issue from 2 weeks ago and the one from 3 days ago is the same but they both involve floating point and both involve failures in tests that shouldn't have been impacted by the subject patch. I will look at the failures seen here: https://github.com/bytecodealliance/wasmtime/runs/1476300496?check_suite_focus=true#step:7:1883 during merging. Try to reproduce and also see what valgrind reports.
abrown commented on Issue #2432:
For reference, previously
build.rs
had some code that would disable certain FP tests when using a certain version of Linux libc that caused FP parsing issues. This was removed here with the switch from wabt to wat. After looking at it, I am discarding this as a cause since we have not seen issues with that in the intervening year (these failures are recent and SIMD-related) but I bring it up since parsing still could be a potential cause.
Last updated: Dec 23 2024 at 12:05 UTC