[DebugInfo] Add DW_AT_byte_size to vectors

[mattd]

Summary

This patch adds the DW_AT_byte_size dwarf attribute to vectors.
This fixes PR21924

LLVM will round a vector up to the next alignable address, which can result in
the vector's representation in the object file being larger than what the
debugger will calculate via NumberOfElements * ElementSize. In such a case calling sizeof(MyVec) in the source will result in a different value than what a debugger might present. This situation can occur because LLVM permits non-power of two 'vector_size' attributes.

[aprantl]

Minor comments inline, but otherwise this seems reasonable. Thanks!

lib/CodeGen/AsmPrinter/DwarfUnit.cpp

1426

Could you add && "not a subrange" and make sure that the Verifier complains when this condition isn't met?

[probinson]

lib/CodeGen/AsmPrinter/DwarfUnit.cpp

1428

I think get<ConstantInt *> instead of dyn_cast so this will assert if it isn't the right type.

[echristo]

lib/CodeGen/AsmPrinter/DwarfUnit.cpp

1428

Yes, because otherwise you could be grabbing a null pointer as well.

[probinson]

lib/CodeGen/AsmPrinter/DwarfUnit.cpp

1428

I'm not really familiar with PointerUnion; the value can be set but still null? So there needs to be an explicit assert here anyway.

[echristo]

lib/CodeGen/AsmPrinter/DwarfUnit.cpp

1428

dyn_cast returns null if the thing isn't of that type.

[mattd]

Thanks for the feedback!

• I made use of the PointerUnion::get() instead of dyn_cast. The former will assert instead of returning null when the item does not exist.
• Added an assertion message when checking that a vector has a single item in its Elements array (an item of type subrange).
• I will add a verifier check and accompanying test for the verifier in a separate patch.

lib/CodeGen/AsmPrinter/DwarfUnit.cpp

   IndexTyDie = &createAndAddDIE(dwarf::DW_TAG_base_type, getUnitDie());
   addString(*IndexTyDie, dwarf::DW_AT_name, "sizetype");
   addUInt(*IndexTyDie, dwarf::DW_AT_byte_size, None, sizeof(int64_t));
   addUInt(*IndexTyDie, dwarf::DW_AT_encoding, dwarf::DW_FORM_data1,
           dwarf::DW_ATE_unsigned);
   return IndexTyDie;
 }
 
+/// Returns true if the vector's size differs from the sum of sizes of elements
+/// the user specified.  This can occur if the vector has been rounded up to
+/// fit memory alignment constraints.
+static bool hasVectorBeenPadded(const DICompositeType *CTy) {
+  assert(CTy && CTy->isVector() && "Composite type is not a vector");
+  const uint64_t ActualSize = CTy->getSizeInBits();
+
+  // Obtain the size of each element in the vector.
+  DIType *BaseTy = CTy->getBaseType().resolve();
+  assert(BaseTy && "Unknown vector element type.");
+  const uint64_t ElementSize = BaseTy->getSizeInBits();
+
+  // Locate the number of elements in the vector.
+  const DINodeArray Elements = CTy->getElements();
+  assert(Elements.size() == 1 &&
+         Elements[0]->getTag() == dwarf::DW_TAG_subrange_type);
+  const auto Subrange = cast<DISubrange>(Elements[0]);
+  const auto CI = Subrange->getCount().dyn_cast<ConstantInt *>();
+  const int32_t NumVecElements = CI->getSExtValue();
+
+  // Ensure we found the element count and that the actual size is wide
+  // enough to contain the requested size.
+  assert(ActualSize >= (NumVecElements * ElementSize) && "Invalid vector size");
+  return ActualSize != (NumVecElements * ElementSize);
+}
+
 void DwarfUnit::constructArrayTypeDIE(DIE &Buffer, const DICompositeType *CTy) {
-  if (CTy->isVector())
+  if (CTy->isVector()) {
     addFlag(Buffer, dwarf::DW_AT_GNU_vector);
+    if (hasVectorBeenPadded(CTy))
+      addUInt(Buffer, dwarf::DW_AT_byte_size, None,
+              CTy->getSizeInBits() / CHAR_BIT);
+  }
 
   // Emit the element type.
   addType(Buffer, resolve(CTy->getBaseType()));
 
   // Get an anonymous type for index type.
   // FIXME: This type should be passed down from the front end
   // as different languages may have different sizes for indexes.
   DIE *IdxTy = getIndexTyDie();

test/DebugInfo/X86/dbg-vector-size.ll

+; RUN: llc < %s -filetype=obj | llvm-dwarfdump -v -debug-info - | FileCheck %s
+;
+; CHECK:      DW_TAG_array_type
+; CHECK-NEXT: DW_AT_GNU_vector [DW_FORM_flag_present] (true)
+; CHECK-NEXT: DW_AT_byte_size [DW_FORM_data1] (0x20)
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+define dso_local void @test() !dbg !9 {
+  %1 = alloca <6 x float>, align 32
+  call void @llvm.dbg.declare(metadata <6 x float>* %1, metadata !13, metadata !DIExpression()), !dbg !19
+  %2 = bitcast <6 x float>* %1 to i8*, !dbg !20
+  call void @foo(i8* %2), !dbg !21
+  ret void, !dbg !22
+}
+
+declare void @llvm.dbg.declare(metadata, metadata, metadata)
+declare dso_local void @foo(i8*)
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!5, !6, !7}
+!llvm.ident = !{!8}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, retainedTypes: !3)
+!1 = !DIFile(filename: "small.c", directory: "/dbg/info")
+!2 = !{}
+!3 = !{!4}
+!4 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: null, size: 64)
+!5 = !{i32 2, !"Dwarf Version", i32 4}
+!6 = !{i32 2, !"Debug Info Version", i32 3}
+!7 = !{i32 1, !"wchar_size", i32 4}
+!8 = !{!"clang version 7.0.0 (trunk 326140)"}
+!9 = distinct !DISubprogram(name: "test", scope: !10, file: !10, line: 3, type: !11, isLocal: false, isDefinition: true, scopeLine: 3, flags: DIFlagPrototyped, isOptimized: false, unit: !0, variables: !2)
+!10 = !DIFile(filename: "./small.c", directory: "/dbg/info")
+!11 = !DISubroutineType(types: !12)
+!12 = !{null}
+!13 = !DILocalVariable(name: "v01", scope: !9, file: !10, line: 4, type: !14)
+!14 = !DIDerivedType(tag: DW_TAG_typedef, name: "vec6f", file: !10, line: 1, baseType: !15)
+!15 = !DICompositeType(tag: DW_TAG_array_type, baseType: !16, size: 256, flags: DIFlagVector, elements: !17)
+!16 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float)
+!17 = !{!18}
+!18 = !DISubrange(count: 6)
+!19 = !DILocation(line: 4, column: 9, scope: !9)
+!20 = !DILocation(line: 5, column: 7, scope: !9)
+!21 = !DILocation(line: 5, column: 3, scope: !9)
+!22 = !DILocation(line: 6, column: 1, scope: !9)