[CIR][CIRGen][Builtin][X86] Lower mm_prefetch

[RiverDave]

Couple of things I have questions about:

1. I duplicated function getIntValueFromConstOp from CIRGenBuiltinAArch64.cpp. I was wondering if that's correct or if there's a place where we can avoid that duplication.

2. For the tests related to mm_prefetch im not sure if it'd be correct to define them in a file eg: sse-builtins.c like it's currently done in the codegen lib.

3. I'm also aware we can emit a call for a PreFetchOp would that be required in this case?

related: #1414 (A PR was also opened but It was not resolved)

[RiverDave]

Okay, been experimenting with the tests for a while. in the test when including:

• flag: -ffreestanding
• header: x86intrin.h

Causes the intrinsic call operation:

cir.llvm.intrinsic "prefetch" {{%.*}} : (!cir.ptr<!void>, !s32i, !s32i, !s32i) -> !void

To transform in a proper prefetch operation:

    cir.prefetch(%2 : !cir.ptr<!void>) locality(0) read loc(#loc6)

I assume we prefer the 2nd. Is that right?