[llvm] [SLP][AMDGPU] Vectorize operands of non-trivially-vectorizable intrinsic calls (PR #189784)

[Alexey Bataev via llvm-commits]

================
@@ -243,6 +243,52 @@ static const int MinScheduleRegionSize = 16;
 /// Maximum allowed number of operands in the PHI nodes.
 static const unsigned MaxPHINumOperands = 128;
 
+/// For instructions that are not trivially vectorizable, try to vectorize thier
+/// operands.
+static Value *getNonTriviallyVectorizableIntrinsicCallOperand(Value *V) {
----------------
alexey-bataev wrote:

I think we can skip most of the internal check and extend this to support vectorization of the arguments of all non-vectorized functions

https://github.com/llvm/llvm-project/pull/189784