[llvm] Make TargetLowering::getValueType() virtual to fix <N x ptr(7)> crash (PR #126642)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,39 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 -passes=loop-vectorize -S < %s | FileCheck %s
+
+; Reduced from a crash, variables added to make things more realistic.
+; This is a roundabout test for TargetLowering::getValueType() returning
+; a reasonable value for <N x p7> instead of asserting.
+define amdgpu_kernel void @_dynamic_pack_simple_dispatch_0_pack_i32(ptr addrspace(1) %.ptr, i64 %0) {
+; CHECK-LABEL: define amdgpu_kernel void @_dynamic_pack_simple_dispatch_0_pack_i32(
+; CHECK-SAME: ptr addrspace(1) [[DOTPTR:%.*]], i64 [[TMP0:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  [[_LR_PH5:.*:]]
+; CHECK-NEXT:    [[DOTRSRC:%.*]] = call ptr addrspace(8) @llvm.amdgcn.make.buffer.rsrc.p1(ptr addrspace(1) [[DOTPTR]], i16 0, i32 -2147483648, i32 159744)
+; CHECK-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(8) [[DOTRSRC]] to ptr addrspace(7)
+; CHECK-NEXT:    br label %[[BB2:.*]]
+; CHECK:       [[BB2]]:
+; CHECK-NEXT:    [[TMP3:%.*]] = phi i64 [ 0, [[DOTLR_PH5:%.*]] ], [ [[TMP5:%.*]], %[[BB2]] ]
+; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr i32, ptr addrspace(7) [[TMP1]], i32 0
+; CHECK-NEXT:    [[TMP5]] = add i64 [[TMP3]], 1
+; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[TMP3]], [[TMP0]]
+; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], [[DOT_CRIT_EDGE_LOOPEXIT:label %.*]], label %[[BB2]]
+; CHECK:       [[__CRIT_EDGE_LOOPEXIT:.*:]]
+; CHECK-NEXT:    ret void
+;
+.lr.ph5:
+  %.rsrc = call ptr addrspace(8) @llvm.amdgcn.make.buffer.rsrc.p1(ptr addrspace(1) %.ptr, i16 0, i32 2147483648, i32 159744)
+  %1 = addrspacecast ptr addrspace(8) %.rsrc to ptr addrspace(7)
----------------
arsenm wrote:

Use named values in tests 

https://github.com/llvm/llvm-project/pull/126642