Fix STLUR for 0 imm

[EgorBo]

Fixes #105906

[EgorBo]

PTAL @dotnet/jit-contrib this unblocks Debug build for NativeAOT on macOS (or any other RCPC2 arm64 hardware).

Some code creates LEA(addr, 0) not sure if we should create such unscaled LEAs with 0 offset and zero index in the first place, but looks like it's valid (in this case the 0 was a field sequence ICON) - this led to stlur with 0 offset. stlur (and it load counterpart - ldapur) always use IF_LS_2C encoding, so we shouldn't use the same path as for other loads/stores.

Generally, we emit ldapur/stlur only for volatile loads/stores with LEA and a small offset (no index, no scale) since these instructions don't support any other addressing mode.

PS: I wasn't able to come up with a test which creates LEA(base + 0) tree, so leaving Debug repo build as a test.

[EgorBo]

PTAL @amanasifkhalid @TIHan as emitter experts

[amanasifkhalid]

I don't know about "emitter expert," but this LGTM.

PS: I wasn't able to come up with a test which creates LEA(base + 0) tree, so leaving Debug repo build as a test.

It might be a good idea to add tests for this pattern to codegenarm64test.cpp. You can then run those tests with DOTNET_JitEmitUnitTests="*" and DOTNET_JitEmitUnitTestsSections="general", and just replay a single method with SPMI.

Thanks!

[amanasifkhalid]

While you're here, could you please move these cases up to INS_ldurb/INS_sturb since they have the same behavior?

[EgorBo]

Addressed. They indeed have the same issue but we never use them 🙂