[PATCH] D121951: [AMDGPU] Only warn when mixing printf and hostcall

[Scott Linder via Phabricator via llvm-commits]

scott.linder created this revision.
Herald added subscribers: foad, kerbowa, hiraditya, t-tye, Anastasia, tpr, dstuttard, yaxunl, nhaehnle, jvesely, kzhuravl, arsenm.
Herald added a project: All.
scott.linder requested review of this revision.
Herald added subscribers: llvm-commits, cfe-commits, wdng.
Herald added projects: clang, LLVM.

We are currently overly conservative, and reject modules which contain
dead uses of hostcall. This has the effect of preventing compilation at
-O0 of a use of printf in an otherwise empty module (i.e. one containing
no use of hostcall).

In a non-single-source world it is not possible to determine whether the
call is dead in the compiler, even if we wanted to.

Instead, make the error a warning, as the compiler cannot diagnose it
without some false positives, but it will remain an issue the user
should be aware of when producing pre-V5 HSA ABI code objects.

Expand testing to modules with printf+no-hostcall, printf+dead-hostcall,
and printf+hostcall to exercise the warning.

Add clang tests to tie the above tests to OpenCL source. Specifically,
the test llvm/test/CodeGen/AMDGPU/opencl-printf-dead-hostcall.ll
represents the case which previously failed at -O0.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D121951

Files:
  clang/test/CodeGenOpenCL/amdgpu-printf.cl
  llvm/lib/Target/AMDGPU/AMDGPUPrintfRuntimeBinding.cpp
  llvm/test/CodeGen/AMDGPU/opencl-printf-and-hostcall.ll
  llvm/test/CodeGen/AMDGPU/opencl-printf-dead-hostcall.ll
  llvm/test/CodeGen/AMDGPU/opencl-printf-no-hostcall.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D121951.416313.patch
Type: text/x-patch
Size: 8548 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220317/332b9852/attachment.bin>