Correctly sequence fgMorphModToSubMulDiv

[CarolEidt]

This method was creating a temp, but the final result was a GT_SUB with
a use of the temp as its op1, and it was not setting GTF_REVERSE_OPS.
This led to a liveness assert in LSRA.

[mikedn]

Oops, I take it this was indirectly caused by my recent change that moved magic division to lowering. But how come this worked for ARM64? Lack of tests?

[CarolEidt]

@mikedn - Yes, the specific scenario needed to reproduce this wasn't covered by existing tests. But at least it was reproducible with a very small test case :-)

[mikedn]

Hmm, I need to take a closer look at this. It seems to me that even with this fix something doesn't work right, if we do arg % 0 we end up with arg, the division by 0 is nowhere to be seen and no exception is thrown. There's actually a bug in my code, it should not call fgMorphModToSubMulDiv if the divisor is 0. But then the ARM64 specific morph code appears to be calling fgMorphModToSubMulDiv unconditionally so presumably it should handle division by 0 correctly.

[mikedn]

I opened a separate issue for the problem I'm seeing, it appears to be completely unrelated to the bug you are fixing here - #8648.

[CarolEidt]

@dotnet/jit-contrib PTAL. This fixes bug 359736

[sivarv]

[briansull]

LGTM