[LegalizeTypes][AMDGPU]: Allow for scalarization of insert_subvector (#…

…104236)

Legalization for when the inserted subvector is to be scalarized.

https://godbolt.org/z/vx3joWqoh

llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h

@@ -865,6 +865,7 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
   SDValue ScalarizeVecOp_UnaryOp(SDNode *N);
   SDValue ScalarizeVecOp_UnaryOp_StrictFP(SDNode *N);
   SDValue ScalarizeVecOp_CONCAT_VECTORS(SDNode *N);
+  SDValue ScalarizeVecOp_INSERT_SUBVECTOR(SDNode *N, unsigned OpNo);
   SDValue ScalarizeVecOp_EXTRACT_VECTOR_ELT(SDNode *N);
   SDValue ScalarizeVecOp_VSELECT(SDNode *N);
   SDValue ScalarizeVecOp_VSETCC(SDNode *N);

llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

@@ -765,6 +765,9 @@ bool DAGTypeLegalizer::ScalarizeVectorOperand(SDNode *N, unsigned OpNo) {
   case ISD::CONCAT_VECTORS:
     Res = ScalarizeVecOp_CONCAT_VECTORS(N);
     break;
+  case ISD::INSERT_SUBVECTOR:
+    Res = ScalarizeVecOp_INSERT_SUBVECTOR(N, OpNo);
+    break;
   case ISD::EXTRACT_VECTOR_ELT:
     Res = ScalarizeVecOp_EXTRACT_VECTOR_ELT(N);
     break;
@@ -882,6 +885,19 @@ SDValue DAGTypeLegalizer::ScalarizeVecOp_CONCAT_VECTORS(SDNode *N) {
   return DAG.getBuildVector(N->getValueType(0), SDLoc(N), Ops);
 }
 
+/// The inserted subvector is to be scalarized - use insert vector element
+/// instead.
+SDValue DAGTypeLegalizer::ScalarizeVecOp_INSERT_SUBVECTOR(SDNode *N,
+                                                          unsigned OpNo) {
+  // We should not be attempting to scalarize the containing vector
+  assert(OpNo == 1);
+  SDValue Elt = GetScalarizedVector(N->getOperand(1));
+  SDValue ContainingVec = N->getOperand(0);
+  return DAG.getNode(ISD::INSERT_VECTOR_ELT, SDLoc(N),
+                     ContainingVec.getValueType(), ContainingVec, Elt,
+                     N->getOperand(2));
+}
+
 /// If the input is a vector that needs to be scalarized, it must be <1 x ty>,
 /// so just return the element, ignoring the index.
 SDValue DAGTypeLegalizer::ScalarizeVecOp_EXTRACT_VECTOR_ELT(SDNode *N) {

llvm/test/CodeGen/AMDGPU/scalarize-insert-subvector.ll

@@ -0,0 +1,24 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 < %s | FileCheck -check-prefix=GCN %s
+
+define void @scalarize_insert_subvector(ptr addrspace(3) %inptr, ptr addrspace(3) %inptr1, ptr addrspace(3) %outptr) {
+; GCN-LABEL: scalarize_insert_subvector:
+; GCN:       ; %bb.0:
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    ds_read_b64 v[4:5], v0
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    ds_read_b32 v5, v1 offset:4
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    ds_write_b64 v2, v[4:5]
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    s_setpc_b64 s[30:31]
+  %load0 = load <2 x i32>, ptr addrspace(3) %inptr, align 8
+  %load1= load <2 x i32>, ptr addrspace(3) %inptr1, align 8
+  %shuffle0 = shufflevector <2 x i32> %load1, <2 x i32> poison, <1 x i32> <i32 1>
+  %bitcast0 = bitcast <1 x i32> %shuffle0 to <2 x half>
+  %bitcast1 = bitcast <2 x i32> %load0 to <4 x half>
+  %shuffle1 = shufflevector <2 x half> %bitcast0, <2 x half> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
+  %shuffle2 = shufflevector <4 x half> %bitcast1, <4 x half> %shuffle1, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
+  store <4 x half> %shuffle2, ptr addrspace(3) %outptr
+  ret void
+}