[DAGCombiner] Use `getAnyExtOrTrunc` instead of TRUNCATE in ExtractVectorElt combine

[Pierre-vh]

Summary

ScalarVT isn't guaranteed to be smaller than the BCSrc.

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

if (IndexC && VecOp.getOpcode() == ISD::BITCAST && VecVT.isInteger() &&
       VecOp.hasOneUse()) {
     // The vector index of the LSBs of the source depend on the endian-ness.
     bool IsLE = DAG.getDataLayout().isLittleEndian();
     unsigned ExtractIndex = IndexC->getZExtValue();
     // extract_elt (v2i32 (bitcast i64:x)), BCTruncElt -> i32 (trunc i64:x)
     unsigned BCTruncElt = IsLE ? 0 : NumElts - 1;
     SDValue BCSrc = VecOp.getOperand(0);
     if (ExtractIndex == BCTruncElt && BCSrc.getValueType().isScalarInteger())
-      return DAG.getNode(ISD::TRUNCATE, DL, ScalarVT, BCSrc);
+      return DAG.getAnyExtOrTrunc(BCSrc, DL, ScalarVT);
 
     if (LegalTypes && BCSrc.getValueType().isInteger() &&
         BCSrc.getOpcode() == ISD::SCALAR_TO_VECTOR) {
       // ext_elt (bitcast (scalar_to_vec i64 X to v2i64) to v4i32), TruncElt -->
       // trunc i64 X to i32
       SDValue X = BCSrc.getOperand(0);
       assert(X.getValueType().isScalarInteger() && ScalarVT.isScalarInteger() &&
              "Extract element and scalar to vector can't change element type "

llvm/test/CodeGen/AMDGPU/dagcombine-v1i8-extractvecelt-crash.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -march=amdgcn -mcpu=gfx908 < %s | FileCheck %s
+
+define void @wombat(i1 %cond, <1 x i8> addrspace(5)* %addr) {
+; CHECK-LABEL: wombat:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    buffer_load_ubyte v2, v1, s[0:3], 0 offen
+; CHECK-NEXT:    v_and_b32_e32 v0, 1, v0
+; CHECK-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; CHECK-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; CHECK-NEXT:    s_cbranch_execz .LBB0_2
+; CHECK-NEXT:  ; %bb.1: ; %then
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    v_mov_b32_e32 v2, 0
+; CHECK-NEXT:  .LBB0_2: ; %end
+; CHECK-NEXT:    s_or_b64 exec, exec, s[4:5]
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    buffer_store_byte v2, v1, s[0:3], 0 offen
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
+entry:
+  %load = load <1 x i8>, <1 x i8> addrspace(5)* %addr, align 1
+  br i1 %cond, label %then, label %end
+
+then:
+  br label %end
+
+end:
+  %phi_value = phi <1 x i8> [%load, %entry], [zeroinitializer, %then]
+  store <1 x i8> %phi_value, <1 x i8> addrspace(5)* %addr, align 1
+  ret void
+}