Fix weird CMake issue with custom LLVM

[LebedevRI]

Without this, cmake fails with:

CMake Error in dependencies/llvm/CMakeLists.txt:
  Target "Halide_LLVM" INTERFACE_INCLUDE_DIRECTORIES property contains path:

    "/repositories/halide/dependencies/llvm/"

  which is prefixed in the source directory.

LLVM_INCLUDE_DIRS there is /repositories/llvm-project/llvm/include;/builddirs/llvm-project/build-Clang13/include,
and INTERFACE_INCLUDE_DIRECTORIES's property beforehand is `` (empty),
but after this line it suddenly becomes /repositories/halide/dependencies/llvm/$<BUILD_INTERFACE:/repositories/llvm-project/llvm/include;/builddirs/llvm-project/build-Clang13/include>.

This is quite obscure. I don't really understand what is going on,
but with the patch it builds fine.

[alexreinking]

Thanks! I hate CMake's quoting rules...

[LebedevRI]

Thanks! I hate CMake's quoting rules...

Note: i really do not understand why quotes are needed here...

[alexreinking]

Thanks! I hate CMake's quoting rules...

Note: i really do not understand why quotes are needed here...

Because it turns it into two arguments...

$<BUILD_INTERFACE:before-semicolon

and

after-semicolon>

And because CMake is inconsistent in how it treats this.

[alexreinking]

Failures are unrelated. Tagging @steven-johnson for the llvm-14 fix.

[LebedevRI]

Failures are unrelated. Tagging @steven-johnson for the llvm-14 fix.

Looks like the CI broke a week ago - https://buildbot.halide-lang.org/master/#/builders/173/builds/499 vs https://buildbot.halide-lang.org/master/#/builders/173/builds/500

cc @labrinea

ada028c32f47ca84a0b7be5d1ab4e3c943f859a3 is the first bad commit
commit ada028c32f47ca84a0b7be5d1ab4e3c943f859a3
Author: Alexandros Lamprineas <alexandros.lamprineas@arm.com>
Date:   Tue Dec 21 15:41:06 2021 +0000

    [AArch64] Add a tablegen pattern for UZP2.
    
    Converts concat_vectors((trunc (lshr)), (trunc (lshr))) to UZP2
    when the shift amount is half the width of the vector element.
    
    Differential Revision: https://reviews.llvm.org/D116021

 llvm/lib/Target/AArch64/AArch64InstrInfo.td     | 13 +++++++
 llvm/test/CodeGen/AArch64/arm64-uzp2-combine.ll | 47 +++++++++++++++++++++++++
 2 files changed, 60 insertions(+)
 create mode 100644 llvm/test/CodeGen/AArch64/arm64-uzp2-combine.ll
$ git bisect log
git bisect start
# bad: [2950a0cbc981e07d319a8cd3f9efa8bdf71b4e5d] z
git bisect bad 2950a0cbc981e07d319a8cd3f9efa8bdf71b4e5d
# good: [ecb9d8e4e3c4623c2edcd2c50727103927d31508] [LICM] Hoist LOAD without sinking the STORE
git bisect good ecb9d8e4e3c4623c2edcd2c50727103927d31508
# good: [78b0f3701d441e887e92ff9feec4c226c67c334f] [HIPSPV][1/4] Refactor HIP tool chain
git bisect good 78b0f3701d441e887e92ff9feec4c226c67c334f
# good: [6969f8415df764a2b267827e2b8e88869668a654] [runtimes] Fix type on flag name in D115852
git bisect good 6969f8415df764a2b267827e2b8e88869668a654
# bad: [ad761f0c3961a09dffd3e7eaea82cd331db9184f] [mlir] Update BUILD.bazel to include `scf_tests`
git bisect bad ad761f0c3961a09dffd3e7eaea82cd331db9184f
# good: [8f85d5205da0310a40fcb92d1830ffa1de99d566] [tsan] Disable test from D115759 on Darwin
git bisect good 8f85d5205da0310a40fcb92d1830ffa1de99d566
# bad: [a3ea9052d6a16b13607046df6a324403fb51888d] [PowerPC] Do not increase cost for getUserCost with MMA types
git bisect bad a3ea9052d6a16b13607046df6a324403fb51888d
# good: [6e28b86cc629c351b1f5d238146af1b6f7ceb785] AlignConsecutiveDeclarations not working for 'const' keyword in JavsScript
git bisect good 6e28b86cc629c351b1f5d238146af1b6f7ceb785
# bad: [00ec441253048f5e30540ea26bb0a28c42a5fc18] [Clang] debug-info-objname.cpp test: explictly encode a x86 target when using %clang_cl to avoid falling back to a native CPU triple.
git bisect bad 00ec441253048f5e30540ea26bb0a28c42a5fc18
# good: [d5b3cb07118ae41542ca2d687d3af0a0f10c0472] [libc++][NFC] Fix links to https://llvm.org/PR20183 in the tests
git bisect good d5b3cb07118ae41542ca2d687d3af0a0f10c0472
# good: [36ea9861e3b58a715434cc2ec5ee2ebf7053bbc9] [clang-format] Remove unnecessary qualifications. NFC.
git bisect good 36ea9861e3b58a715434cc2ec5ee2ebf7053bbc9
# bad: [ada028c32f47ca84a0b7be5d1ab4e3c943f859a3] [AArch64] Add a tablegen pattern for UZP2.
git bisect bad ada028c32f47ca84a0b7be5d1ab4e3c943f859a3
# good: [1a929525e86a20d0b3455a400d0dbed40b325a13] [clangd] Return error for textdocument/outgoingCalls rather than success
git bisect good 1a929525e86a20d0b3455a400d0dbed40b325a13
# first bad commit: [ada028c32f47ca84a0b7be5d1ab4e3c943f859a3] [AArch64] Add a tablegen pattern for UZP2.

[LebedevRI]

Ok, i think there is no LLVM-side regression, we just need to look for the different instruction now, posted #6521.
Ok, a better snippet does suggest it's a LLVM regression: https://godbolt.org/z/EeqjGW58x

[alexreinking]

I opened an issue with LLVM here: llvm/llvm-project#52919

[LebedevRI]

@alexreinking thank you!