[PATCH] D134463: [AMDGPU] Use V_PERM to match buildvectors when inputs are not canonicalized (i.e. can't use V_PACK)

[Jeffrey Byrnes via Phabricator via llvm-commits]

jrbyrnes created this revision.
Herald added subscribers: kosarev, foad, kerbowa, hiraditya, t-tye, tpr, dstuttard, yaxunl, jvesely, kzhuravl, arsenm.
Herald added a project: All.
jrbyrnes requested review of this revision.
Herald added subscribers: llvm-commits, wdng.
Herald added a project: LLVM.

If we can not prove that f16 operands of a buildvector are canonicalized, then we can not lower into a V_PACK. In this scenario, we would previously lower into some combination of and(sdwa), shr, or. This patch allows for matching into V_PERM instead -- which uses SGPR instead of VGPR (or encodes the literal in the instruction itself), and has less VALU latency.

Change-Id: Ifa4a74fdb81ef44f22ba490c7fdf81ec8aebc945


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D134463

Files:
  llvm/lib/Target/AMDGPU/SIInstructions.td
  llvm/test/CodeGen/AMDGPU/build-vector-packed-partial-undef.ll
  llvm/test/CodeGen/AMDGPU/extract-subvector-16bit.ll
  llvm/test/CodeGen/AMDGPU/insert_vector_elt.v2i16.ll
  llvm/test/CodeGen/AMDGPU/pack.v2f16.ll
  llvm/test/CodeGen/AMDGPU/v_perm_non_canon.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D134463.462251.patch
Type: text/x-patch
Size: 18087 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220922/7dfc4042/attachment-0001.bin>