[DebugInfo] Remove the restriction on the size of DIStringType in DebugHandlerBase::isUnsignedDIType

[cchen15]

Summary

It turns out that the optimizations may correctly encode a Fortran character object as an integer even when the character length is greater than 1.

Also, it is probably not the AsmPrinter's job to catch the result of a bad transformation, which the removed assertion appears to be doing.

This is a follow-up to https://reviews.llvm.org/D100015.

[aprantl]

Seems reasonable.

llvm/test/CodeGen/Generic/dbg-distringtype-uint.ll

67

Can you remove all attributes that are not strictly necessary (especially the ones in quotes)?

[cchen15]

llvm/test/CodeGen/Generic/dbg-distringtype-uint.ll

67

Excessive attributes eliminated. Thanks for the review, Adrian!

[cchen15]

Hi @aprantl. I just got notified by buildbots that the new test fails on clang-aarch64-quick and clang-armv7-quick (maybe more to come). Could you please suggest a way to disable this test on triples that don't support the 'x86_64-unknown-linux-gnu' target? Thanks!

[shchenz]

I think the reason is you specify triple for a case in target-independent dir(test/CodeGen/Generic), you need to add a "REQUIRES:" in the test file to make sure the test is only run on buildbots that supports X86-linux. See test/CodeGen/Generic/csw-debug-assert.ll.

[aprantl]

In D108559#2973690, @shchenz wrote:

I think the reason is you specify triple for a case in target-independent dir(test/CodeGen/Generic), you need to add a "REQUIRES:" in the test file to make sure the test is only run on buildbots that supports X86-linux. See test/CodeGen/Generic/csw-debug-assert.ll.

Even better is to move the test to the x86 subdirectory where the lit.cfg automatically adds this requirement to all tests.

[cchen15]

Issue fixed as suggested in https://reviews.llvm.org/D108946.

llvm/lib/CodeGen/AsmPrinter/DebugHandlerBase.cpp

uint64_t DebugHandlerBase::getBaseTypeSize(const DIType *Ty) {
   if (BaseType->getTag() == dwarf::DW_TAG_reference_type ||
       BaseType->getTag() == dwarf::DW_TAG_rvalue_reference_type)
     return Ty->getSizeInBits();
 
   return getBaseTypeSize(BaseType);
 }
 
 bool DebugHandlerBase::isUnsignedDIType(const DIType *Ty) {
-  // SROA may generate dbg value intrinsics to assign an unsigned value to a
-  // Fortran CHARACTER(1) type variables. Make them as unsigned.
   if (isa<DIStringType>(Ty)) {
-    assert((Ty->getSizeInBits()) == 8 && "Not a valid unsigned type!");
+    // Some transformations (e.g. instcombine) may decide to turn a Fortran
+    // character object into an integer, and later ones (e.g. SROA) may
+    // further inject a constant integer in a llvm.dbg.value call to track
+    // the object's value. Here we trust the transformations are doing the
+    // right thing, and treat the constant as unsigned to preserve that value
+    // (i.e. avoid sign extension).
     return true;
   }
+
   if (auto *CTy = dyn_cast<DICompositeType>(Ty)) {
     // FIXME: Enums without a fixed underlying type have unknown signedness
     // here, leading to incorrectly emitted constants.
     if (CTy->getTag() == dwarf::DW_TAG_enumeration_type)
       return false;
 
     // (Pieces of) aggregate types that get hacked apart by SROA may be
     // represented by a constant. Encode them as unsigned bytes.

llvm/test/CodeGen/Generic/dbg-distringtype-uint.ll

+; RUN: llc -filetype=obj < %s | llvm-dwarfdump -debug-info - | FileCheck %s
+;
+; CHECK: [[SYM:[a-z0-9]+]]:  DW_TAG_formal_parameter
+; CHECK:                     DW_AT_name ("esym")
+; CHECK:                     DW_AT_type ([[TYPE:[a-z0-9]+]] "CHARACTER_1")
+;
+; CHECK:                     DW_TAG_formal_parameter
+; CHECK:                       DW_AT_const_value  (7523094288207667809)
+; CHECK:                       DW_AT_abstract_origin  ([[SYM]] "esym")
+;
+; CHECK: [[TYPE]]:           DW_TAG_string_type
+; CHECK:                       DW_AT_name ("CHARACTER_1")
+; CHECK:                       DW_AT_byte_size  (0x08)
+;
+; The following IR is obtained by compiling the following Fortran
+; program with -O2 -g (with irrelevant instructions and metadata
+; trimmed):
+;
+; module semiempirical_corrections
+;
+;    implicit none
+;
+; contains
+; subroutine gcpcor(n,iz)
+;    implicit none
+;    integer :: n, i
+;    integer :: iz(n)
+;    iz(i)=1
+;    print*, esym(iz(i))
+; end subroutine
+;
+;
+; character*8 function esym(i)
+;    integer :: i
+;    character*8 elemnt(1)
+;    data elemnt/'abcdefgh'/
+;    esym=elemnt(i)
+;    return
+; end function
+; end module
+;
+; The optimizations encode the constant string as an i64 constant in
+; the debug info for esym, the return variable of the function esym.
+;
+; ModuleID = 'se6.bc'
+source_filename = "se6.f90"
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+@"semiempirical_corrections_mp_esym_$ELEMNT" = internal unnamed_addr constant [1 x [8 x i8]] [[8 x i8] c"abcdefgh"], align 8, !dbg !0
+
+; Function Attrs: nofree nounwind uwtable
+define void @semiempirical_corrections_mp_gcpcor_() local_unnamed_addr #1 !dbg !22 {
+alloca_1:
+  %"var$47" = alloca i64, align 8, !dbg !34
+  call void @llvm.dbg.value(metadata i64 7523094288207667809, metadata !12, metadata !DIExpression()), !dbg !40
+  store i64 7523094288207667809, i64* %"var$47", align 8, !dbg !41, !alias.scope !42, !noalias !45
+  ret void, !dbg !48
+}
+
+; Function Attrs: nofree nosync nounwind readnone speculatable willreturn
+declare void @llvm.dbg.declare(metadata, metadata, metadata) #3
+
+; Function Attrs: nofree nosync nounwind readnone speculatable willreturn
+declare void @llvm.dbg.value(metadata, metadata, metadata) #3
+
+attributes #0 = { mustprogress nofree norecurse nosync nounwind readnone uwtable willreturn "frame-pointer"="none" "intel-lang"="fortran" "min-legal-vector-width"="0" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" }
+attributes #1 = { nofree nounwind uwtable "frame-pointer"="none" "intel-lang"="fortran" "min-legal-vector-width"="0" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" }
+attributes #2 = { mustprogress nofree norecurse nosync nounwind uwtable willreturn "frame-pointer"="none" "intel-lang"="fortran" "min-legal-vector-width"="0" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" }
+attributes #3 = { nofree nosync nounwind readnone speculatable willreturn }
+attributes #4 = { nofree "intel-lang"="fortran" }
+attributes #5 = { nounwind }
+
+!llvm.module.flags = !{!20, !21}
+!llvm.dbg.cu = !{!8}
+!omp_offload.info = !{}
+
+!0 = !DIGlobalVariableExpression(var: !1, expr: !DIExpression())
+!1 = distinct !DIGlobalVariable(name: "elemnt", linkageName: "semiempirical_corrections_mp_esym_$ELEMNT", scope: !2, file: !3, line: 17, type: !16, isLocal: true, isDefinition: true)
+!2 = distinct !DISubprogram(name: "esym", linkageName: "semiempirical_corrections_mp_esym_", scope: !4, file: !3, line: 15, type: !5, scopeLine: 15, spFlags: DISPFlagDefinition, unit: !8, retainedNodes: !11)
+!3 = !DIFile(filename: "se6.f90", directory: "/iusers/cchen15/examples/tests/jr30349/gamess-dga-master/object")
+!4 = !DIModule(scope: null, name: "semiempirical_corrections", file: !3, line: 1)
+!5 = !DISubroutineType(types: !6)
+!6 = !{!7}
+!7 = !DIStringType(name: "CHARACTER_0", size: 64)
+!8 = distinct !DICompileUnit(language: DW_LANG_Fortran95, file: !3, producer: "Intel(R) Fortran 21.0-2745", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !9, globals: !10, splitDebugInlining: false, nameTableKind: None)
+!9 = !{}
+!10 = !{!0}
+!11 = !{!12, !14}
+!12 = !DILocalVariable(name: "esym", arg: 1, scope: !2, file: !3, line: 15, type: !13, flags: DIFlagArtificial)
+!13 = !DIStringType(name: "CHARACTER_1", size: 64)
+!14 = !DILocalVariable(name: "i", arg: 2, scope: !2, file: !3, line: 15, type: !15)
+!15 = !DIBasicType(name: "INTEGER*4", size: 32, encoding: DW_ATE_signed)
+!16 = !DICompositeType(tag: DW_TAG_array_type, baseType: !17, elements: !18)
+!17 = !DIStringType(name: "CHARACTER_2", size: 64)
+!18 = !{!19}
+!19 = !DISubrange(count: 1, lowerBound: 1)
+!20 = !{i32 2, !"Debug Info Version", i32 3}
+!21 = !{i32 2, !"Dwarf Version", i32 4}
+!22 = distinct !DISubprogram(name: "gcpcor", linkageName: "semiempirical_corrections_mp_gcpcor_", scope: !4, file: !3, line: 6, type: !23, scopeLine: 6, spFlags: DISPFlagDefinition, unit: !8)
+!23 = !DISubroutineType(types: !24)
+!24 = !{null}
+!34 = !DILocation(line: 6, column: 19, scope: !22)
+!38 = !DILocation(line: 15, column: 27, scope: !2, inlinedAt: !39)
+!39 = distinct !DILocation(line: 11, column: 13, scope: !22)
+!40 = !DILocation(line: 0, scope: !2, inlinedAt: !39)
+!41 = !DILocation(line: 19, column: 5, scope: !2, inlinedAt: !39)
+!42 = !{!43}
+!43 = distinct !{!43, !44, !"semiempirical_corrections_mp_esym_: %semiempirical_corrections_mp_esym_$ESYM"}
+!44 = distinct !{!44, !"semiempirical_corrections_mp_esym_"}
+!45 = !{!46}
+!46 = distinct !{!46, !44, !"semiempirical_corrections_mp_esym_: %I"}
+!48 = !DILocation(line: 12, column: 1, scope: !22)