[llvm-branch-commits] [llvm] AMDGPU/GlobalISel: AMDGPURegBankLegalize (PR #112864)

[Matt Arsenault via llvm-branch-commits]

================
@@ -106,3 +106,59 @@ void IntrinsicLaneMaskAnalyzer::findLCSSAPhi(Register Reg) {
       S32S64LaneMask.insert(LCSSAPhi.getOperand(0).getReg());
   }
 }
+
+static LLT getReadAnyLaneSplitTy(LLT Ty) {
+  if (Ty.isVector()) {
+    LLT ElTy = Ty.getElementType();
+    if (ElTy == LLT::scalar(16))
+      return LLT::fixed_vector(2, 16);
+    // S32, S64 or pointer
+    return ElTy;
+  }
+
+  // Large scalars and 64-bit pointers
+  return LLT::scalar(32);
+}
+
+static Register buildReadAnyLane(MachineIRBuilder &B, Register VgprSrc,
----------------
arsenm wrote:

Move the function body to avoid forward declaring 

https://github.com/llvm/llvm-project/pull/112864