[PATCH] D135145: [AMDGPU][GISel] Combine G_INSERT_VECTOR_ELT to G_SHUFFLE_VECTOR

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUCombine.td:48-52
+def insert_vec_elt_to_shuffle : GICombineRule<
+  (defs root:$insertelt, unsigned_matchinfo:$matchinfo),
+  (match (wip_match_opcode G_INSERT_VECTOR_ELT):$insertelt,
+      [{ return PreLegalizerHelper.matchInsertVectorEltToShuffle(*${insertelt}, ${matchinfo}); }]),
+  (apply [{ PreLegalizerHelper.applyInsertVectorEltToShuffle(*${insertelt}, ${matchinfo}); }])>;
----------------
This is generic


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUPreLegalizerCombiner.cpp:162-164
+  // This is only beneficial if scalar pack insts are present.
+  if (!MI.getMF()->getSubtarget<GCNSubtarget>().hasScalarPackInsts())
+    return false;
----------------
The DAG treats this as an initial canonicalization, so the obvious codegen benefit isn't so important


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUPreLegalizerCombiner.cpp:166-170
+  // TODO: Only on small vectors?
+  LLT VecTy = MRI.getType(MI.getOperand(0).getReg());
+  if (VecTy.getElementType() != LLT::scalar(16) ||
+      (VecTy.getSizeInBits() % 32) != 0)
+    return false;
----------------
Don't see why this would restrict the vector type


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D135145/new/

https://reviews.llvm.org/D135145