[debug-info] Emit DW_AT_byte_size for non-default pointer/reference types

[arichardson]

Summary

In the CHERI and Arm Morello targets we support a compilation mode that
has multiple pointer sizes (32/64-bit integers as well as capabilities that are
twice that size). In order to support this in the debugger we emit
DW_AT_byte_size for pointer and reference types if their size is different
from the target's pointer size for the alloca address space.

This patch found the pointer size mismatch that I fixed in D99410.

[arichardson]

• update test

[dblaikie]

Sounds OK to me - I'm not 100% sure that's the right address size check to do, but if it works for you, sounds OK to me.

[jrtc27]

In D99411#2856131, @dblaikie wrote:

Sounds OK to me - I'm not 100% sure that's the right address size check to do, but if it works for you, sounds OK to me.

Yeah, LLVM suffers from having lots of granularity with no notion of a default address space... for us they're all (program, alloca, globals) the same

llvm/lib/CodeGen/AsmPrinter/DwarfUnit.cpp

771–777

Hm, this doesn't really accurately reflect what's happening, but the same is true of the existing comment... maybe something like this?

llvm/lib/CodeGen/AsmPrinter/DwarfUnit.cpp

void DwarfUnit::constructTypeDIE(DIE &Buffer, const DIDerivedType *DTy) {
   // attribute in DW_TAG_typedef DIE.
   if (Tag == dwarf::DW_TAG_typedef && DD->getDwarfVersion() >= 5) {
     uint32_t AlignInBytes = DTy->getAlignInBytes();
     if (AlignInBytes > 0)
       addUInt(Buffer, dwarf::DW_AT_alignment, dwarf::DW_FORM_udata,
               AlignInBytes);
   }
 
-  // Add size if non-zero (derived types might be zero-sized.)
-  if (Size && Tag != dwarf::DW_TAG_pointer_type
-           && Tag != dwarf::DW_TAG_ptr_to_member_type
-           && Tag != dwarf::DW_TAG_reference_type
-           && Tag != dwarf::DW_TAG_rvalue_reference_type)
+  // Add size if non-zero or non-default (derived types might be zero-sized.)
+  if (Size && ((Tag != dwarf::DW_TAG_pointer_type &&
+                Tag != dwarf::DW_TAG_ptr_to_member_type &&
+                Tag != dwarf::DW_TAG_reference_type &&
+                Tag != dwarf::DW_TAG_rvalue_reference_type) ||
+               Size != Asm->getDataLayout().getPointerSize(
+                           Asm->getDataLayout().getAllocaAddrSpace())))
     addUInt(Buffer, dwarf::DW_AT_byte_size, None, Size);
 
   if (Tag == dwarf::DW_TAG_ptr_to_member_type)
     addDIEEntry(Buffer, dwarf::DW_AT_containing_type,
                 *getOrCreateTypeDIE(cast<DIDerivedType>(DTy)->getClassType()));
   // Add source line info if available and TyDesc is not a forward declaration.
   if (!DTy->isForwardDecl())
     addSourceLine(Buffer, DTy);

llvm/test/DebugInfo/Generic/ptrsize-nonstandard.ll

+; REQUIRES: object-emission
+
 ; RUN: %llc_dwarf -O0 -filetype=obj < %s > %t
-; RUN: llvm-dwarfdump -v %t | FileCheck %s
+; RUN: llvm-dwarfdump %t | FileCheck %s
 
-; Check that pointers and references get emitted without size information in
-; DWARF, even if they are so specified in the IR
+; Check that pointers and references with size different to the target's pointer
+; size are emitted with size information in DWARF.
 
-; CHECK: 0x[[O1:[0-9a-f]+]]:   DW_TAG_pointer_type
+; CHECK: 0x{{[0-9a-f]+}}:   DW_TAG_pointer_type
 ; CHECK-NEXT: DW_AT_type [DW_FORM_ref4]
-; CHECK-NOT: DW_AT_byte_size
-; CHECK: 0x[[O2:[0-9a-f]+]]:   DW_TAG_
+; CHECK-NEXT: DW_AT_byte_size [DW_FORM_data1] (0x10)
+; CHECK-NOT: DW_TAG_pointer_type
 
-; CHECK: 0x[[O3:[0-9a-f]+]]:   DW_TAG_reference_type
+; CHECK: 0x{{[0-9a-f]+}}:   DW_TAG_reference_type
 ; CHECK-NEXT: DW_AT_type [DW_FORM_ref4]
-; CHECK-NOT: DW_AT_byte_size
+; CHECK-NEXT: DW_AT_byte_size [DW_FORM_data1] (0x10)
+; CHECK-NOT: DW_TAG_reference_type
 
 define i32 @foo() !dbg !4 {
 entry:
   ret i32 0, !dbg !13
 }
 
 define i32 @bar() !dbg !5 {
 entry:
   ret i32 0, !dbg !16
 }
 
 !llvm.dbg.cu = !{!0}
 !llvm.module.flags = !{!11, !12}
 
 !0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, retainedTypes: !2, globals: !2, imports: !2)
 !1 = !DIFile(filename: "dwarf-test.c", directory: "test")
 !2 = !{}
-!4 = distinct !DISubprogram(name: "foo", scope: !0, file: !1, line: 6, type: !6, isLocal: false, isDefinition: true, scopeLine: 6, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: false, unit: !0, retainedNodes: !2)
-!5 = distinct !DISubprogram(name: "bar", scope: !0, file: !1, line: 6, type: !15, isLocal: false, isDefinition: true, scopeLine: 6, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: false, unit: !0, retainedNodes: !2)
+!4 = distinct !DISubprogram(name: "foo", scope: !0, file: !1, line: 6, type: !6, isLocal: false, isDefinition: true, scopeLine: 6, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: false, unit: !0, variables: !2)
+!5 = distinct !DISubprogram(name: "bar", scope: !0, file: !1, line: 6, type: !15, isLocal: false, isDefinition: true, scopeLine: 6, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: false, unit: !0, variables: !2)
 !6 = !DISubroutineType(types: !7)
 !7 = !{!9}
 !8 = !DIBasicType(name: "int", size: 32, align: 32, encoding: DW_ATE_signed)
-!9 = !DIDerivedType(tag: DW_TAG_pointer_type, scope: !0, baseType: !8, size: 64, align: 64)
-!10 = !DIDerivedType(tag: DW_TAG_reference_type, scope: !0, baseType: !8, size: 64, align: 64)
+!9 = !DIDerivedType(tag: DW_TAG_pointer_type, scope: !0, baseType: !8, size: 128, align: 128)
+!10 = !DIDerivedType(tag: DW_TAG_reference_type, scope: !0, baseType: !8, size: 128, align: 128)
 !11 = !{i32 2, !"Dwarf Version", i32 3}
 !12 = !{i32 1, !"Debug Info Version", i32 3}
 !13 = !DILocation(line: 7, scope: !4)
 !14 = !{!10}
 !15 = !DISubroutineType(types: !14)
 !16 = !DILocation(line: 7, scope: !5)

llvm/test/DebugInfo/Generic/ptrsize.ll

 ; RUN: %llc_dwarf -O0 -filetype=obj < %s > %t
 ; RUN: llvm-dwarfdump -v %t | FileCheck %s
 
-; Check that pointers and references get emitted without size information in
-; DWARF, even if they are so specified in the IR
+; Check that pointers and references with size equal to the target's pointer
+; size get emitted without size information in DWARF, even if they are so
+; specified in the IR.
 
 ; CHECK: 0x[[O1:[0-9a-f]+]]:   DW_TAG_pointer_type
 ; CHECK-NEXT: DW_AT_type [DW_FORM_ref4]
 ; CHECK-NOT: DW_AT_byte_size
 ; CHECK: 0x[[O2:[0-9a-f]+]]:   DW_TAG_
 
 ; CHECK: 0x[[O3:[0-9a-f]+]]:   DW_TAG_reference_type
 ; CHECK-NEXT: DW_AT_type [DW_FORM_ref4]