runtime: Intermittent failures in Fma test

I have seen a couple of failures in this test that I can’t repro, even with various combinations of the COMPlus_Enable flags for the various ISAs. Not sure what’s going on, but there’s no way that changing an ARM-only #define should cause this to fail. See https://ci.dot.net/job/dotnet_coreclr/job/master/job/x86_release_windows_nt_innerloop_prtest/4371/ and also https://ci.dot.net/job/dotnet_coreclr/job/master/job/x86_release_windows_nt_innerloop_prtest/4376/

The impacted test is Windows_NT.x86.Release\JIT\HardwareIntrinsics\X86\Fma_Vector256\Fma_ro\Fma_ro.cmd

It seems to pass on retry.

About this issue

Original URL
State: closed
Created 6 years ago
Comments: 15 (15 by maintainers)

Most upvoted comments

Random numbers are beneficial because they are random.

Well, if you say so…

The JIT is large and complex, there is no guarantee that the managed code I write will definitely cause emitIns_SIMD_R_R_AM to be hit (for example).

Except that most of the JIT doesn’t care about intrinsics so in general there should be no problem to generate IL code that hits the right paths. And if the JIT did care about intrinsics (e.g. optimize them) then tests would not help too much because there would far too many cases to consider.

mikedn on Jun 14, 2018