[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, #1404 (A PR was previously 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 latter. Is that right?

[RiverDave]

cc. @bcardosolopes

[bcardosolopes]

@RiverDave sorry, somehow I missed this.

I assume we prefer the latter. Is that right?

cir.prefetch is preferred. The question is whether it leads to the same LLVM IR output without differences and if we need to change cir.prefetch to accommodate. Have you find any issues while trying to do so?

[RiverDave]

@RiverDave sorry, somehow I missed this.

I assume we prefer the latter. Is that right?

cir.prefetch is preferred. The question is whether it leads to the same LLVM IR output without differences and if we need to change cir.prefetch to accommodate. Have you find any issues while trying to do so?

Two things:

1. I've revised this PR and moved it's test case to its respective test file as in OG and included the headers that again, was present in OG.

2. They seem to be equivalent indeed:

raw intrinsic call:

  cir.func dso_local @test_mm_prefetch(%arg0: !cir.ptr<!s8i> loc(fused[#loc3, #loc4])) extra(#fn_attr) {
    %0 = cir.alloca !cir.ptr<!s8i>, !cir.ptr<!cir.ptr<!s8i>>, ["p", init] {alignment = 8 : i64} loc(#loc10)
    cir.store %arg0, %0 : !cir.ptr<!s8i>, !cir.ptr<!cir.ptr<!s8i>> loc(#loc5)
    %1 = cir.load align(8) %0 : !cir.ptr<!cir.ptr<!s8i>>, !cir.ptr<!s8i> loc(#loc6)
    %2 = cir.const #cir.int<0> : !s32i loc(#loc7)
    %3 = cir.cast(bitcast, %1 : !cir.ptr<!s8i>), !cir.ptr<!void> loc(#loc6)
    %4 = cir.const #cir.int<0> : !s32i loc(#loc8)
    %5 = cir.const #cir.int<0> : !s32i loc(#loc8)
    %6 = cir.const #cir.int<1> : !s32i loc(#loc8)
    %7 = cir.llvm.intrinsic "prefetch" %3, %4, %5, %6 : (!cir.ptr<!void>, !s32i, !s32i, !s32i) -> !void loc(#loc8)
    cir.return loc(#loc2)
  } loc(#loc9)

PrefetchOp:

  cir.func dso_local @_Z16test_mm_prefetchPKc(%arg0: !cir.ptr<!u8i> loc(fused[#loc3, #loc4])) extra(#fn_attr) {
    %0 = cir.alloca !cir.ptr<!u8i>, !cir.ptr<!cir.ptr<!u8i>>, ["p", init] {alignment = 8 : i64} loc(#loc8)
    cir.store %arg0, %0 : !cir.ptr<!u8i>, !cir.ptr<!cir.ptr<!u8i>> loc(#loc5)
    %1 = cir.load align(8) %0 : !cir.ptr<!cir.ptr<!u8i>>, !cir.ptr<!u8i> loc(#loc6)
    %2 = cir.cast(bitcast, %1 : !cir.ptr<!u8i>), !cir.ptr<!void> loc(#loc6)
    cir.prefetch(%2 : !cir.ptr<!void>) locality(0) read loc(#loc6)
    cir.return loc(#loc2)
  } loc(#loc7)

I have one question though, Not just in this, but in other PR's, I've noted functions that have some trailing/leading characters as shown: _Z16test_mm_prefetchPKc what is the reason behind this? why is this not presented without the header I included?

[bcardosolopes]

I've noted functions that have some trailing/leading characters as shown: _Z16test_mm_prefetchPKc what is the reason behind this?

It only happens in C++ - due to C++ mangling. If you cat the file to c++filt -n you gonna get the "original" names.