[all-commits] [llvm/llvm-project] ab6b48: DAG: Avoid stack lowering if bitcast has an illega...

[Matt Arsenault via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: ab6b48b7116d118131e7ef459e4d234b4366564f
      https://github.com/llvm/llvm-project/commit/ab6b48b7116d118131e7ef459e4d234b4366564f
  Author: Matt Arsenault <Matthew.Arsenault at amd.com>
  Date:   2023-01-15 (Sun, 15 Jan 2023)

  Changed paths:
    M llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
    M llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.ll
    M llvm/test/CodeGen/AMDGPU/copy-illegal-type.ll
    M llvm/test/CodeGen/AMDGPU/copy-to-reg-scc-clobber.ll
    M llvm/test/CodeGen/AMDGPU/flat_atomics_i64.ll
    M llvm/test/CodeGen/AMDGPU/global_atomics_i64.ll
    M llvm/test/CodeGen/AMDGPU/insert_vector_dynelt.ll
    M llvm/test/CodeGen/AMDGPU/select.f16.ll
    M llvm/test/CodeGen/AMDGPU/v_madak_f16.ll
    M llvm/test/CodeGen/AMDGPU/vector_shuffle.packed.ll

  Log Message:
  -----------
  DAG: Avoid stack lowering if bitcast has an illegal vector result type

A bitcast of <10 x i32> to <5 x i64> was ending up on the
stack. Instead of doing that, handle the case where the new type
doesn't evenly divide but the elements do. Extract the individual
elements and pad with undef.

Avoids stack usage for bitcasts involving <5 x i64>. In some of these
cases, later optimizations actually eliminated the stack objects but
left behind the unused temporary stack object to final emission.

Fixes: SWDEV-377548