[XCOFF] make .file directive have directory info

[shchenz]

Summary

The .file directive is changed to only have basename in D36018 for ELF.

But on AIX, we require the .file directive to also contain the directory info. This aligns with other AIX compiler like XLC and is required by some AIX tool like DBX.

[shchenz]

Lint warning fix

[hubert.reinterpretcast]

From what I understand, this is meant to fix a situation where dbx is unhappy when the file is not simply compiled in the directory where it resides (using its basename). However, this unhappiness extends even to the compiled-using-basename case when the filename contains "special characters".

For example, a file with a tab character ends up with backslash escapes in the assembly (but the assembler on AIX doesn't interpret these):

$ clang -mllvm -dwarf-inlined-strings=Enable a\    q.c -S -o - | grep -F .file
        .file   "a\tq.c"

[shchenz]

In D99785#2671112, @hubert.reinterpretcast wrote:

From what I understand, this is meant to fix a situation where dbx is unhappy when the file is not simply compiled in the directory where it resides (using its basename). However, this unhappiness extends even to the compiled-using-basename case when the filename contains "special characters".

For example, a file with a tab character ends up with backslash escapes in the assembly (but the assembler on AIX doesn't interpret these):

$ clang -mllvm -dwarf-inlined-strings=Enable a\    q.c -S -o - | grep -F .file
        .file   "a\tq.c"

@hubert.reinterpretcast thanks for providing the example. Yes, it is indeed an issue for the same name in symbol table and in dwinfo section.
As you can see in link https://www.ibm.com/docs/en/aix/7.2?topic=files-file-naming-conventions , special characters are not recommmended to be used in the file name, so can we just ignore the special characters and only handle the directory info in this patch? Thanks.

[hubert.reinterpretcast]

In D99785#2715170, @shchenz wrote:

@hubert.reinterpretcast thanks for providing the example. Yes, it is indeed an issue for the same name in symbol table and in dwinfo section.
As you can see in link https://www.ibm.com/docs/en/aix/7.2?topic=files-file-naming-conventions , special characters are not recommmended to be used in the file name, so can we just ignore the special characters and only handle the directory info in this patch? Thanks.

The file name conventions on AIX does not (and probably cannot) recommend against the use of "national characters". The character escape mechanism being applied is incorrect. The string constants in AIX assembly are mostly-raw strings, with double quotation marks in the string content being immediately repeated to indicate that the end-of-string has not been reached.

[hubert.reinterpretcast]

llvm/test/CodeGen/PowerPC/aix-filename.ll

8

There should be an additional test to check against the desired behaviour observed with relative paths (e.g., ../relative/path/to/file).

[shchenz]

1: add a relative path case

[shchenz]

In D99785#2715798, @hubert.reinterpretcast wrote:

In D99785#2715170, @shchenz wrote:

@hubert.reinterpretcast thanks for providing the example. Yes, it is indeed an issue for the same name in symbol table and in dwinfo section.
As you can see in link https://www.ibm.com/docs/en/aix/7.2?topic=files-file-naming-conventions , special characters are not recommmended to be used in the file name, so can we just ignore the special characters and only handle the directory info in this patch? Thanks.

The file name conventions on AIX does not (and probably cannot) recommend against the use of "national characters". The character escape mechanism being applied is incorrect. The string constants in AIX assembly are mostly-raw strings, with double quotation marks in the string content being immediately repeated to indicate that the end-of-string has not been reached.

Thanks, the special characters will be handled in patch D101280

[hubert.reinterpretcast]

LGTM; thanks!

llvm/include/llvm/MC/MCAsmInfo.h

protected:
   /// True is .comm's and .lcomms optional alignment is to be specified in bytes
   /// instead of log2(n).  Defaults to true.
   bool COMMDirectiveAlignmentIsInBytes = true;
 
   /// Describes if the .lcomm directive for the target supports an alignment
   /// argument and how it is interpreted.  Defaults to NoAlignment.
   LCOMM::LCOMMType LCOMMDirectiveAlignmentType = LCOMM::NoAlignment;
 
+  /// True if the target only has basename for .file directive. False if the
+  /// target also needs the directory along with the basename. Default to true.
+  bool HasBasenameOnlyForFileDirective = true;
+
   // True if the target allows .align directives on functions. This is true for
   // most targets, so defaults to true.
   bool HasFunctionAlignment = true;
 
   /// True if the target has .type and .size directives, this is true for most
   /// ELF targets.  Defaults to true.
   bool HasDotTypeDotSizeDirective = true;
 
public:
   bool getCOMMDirectiveAlignmentIsInBytes() const {
     return COMMDirectiveAlignmentIsInBytes;
   }
 
   LCOMM::LCOMMType getLCOMMDirectiveAlignmentType() const {
     return LCOMMDirectiveAlignmentType;
   }
 
+  bool hasBasenameOnlyForFileDirective() const {
+    return HasBasenameOnlyForFileDirective;
+  }
   bool hasFunctionAlignment() const { return HasFunctionAlignment; }
   bool hasDotTypeDotSizeDirective() const { return HasDotTypeDotSizeDirective; }
   bool hasSingleParameterDotFile() const { return HasSingleParameterDotFile; }
   bool hasIdentDirective() const { return HasIdentDirective; }
   bool hasNoDeadStrip() const { return HasNoDeadStrip; }
   bool hasAltEntry() const { return HasAltEntry; }
   const char *getWeakDirective() const { return WeakDirective; }
   const char *getWeakRefDirective() const { return WeakRefDirective; }

llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp

bool AsmPrinter::doInitialization(Module &M) {
 
   // Allow the target to emit any magic that it wants at the start of the file.
   emitStartOfAsmFile(M);
 
   // Very minimal debug info. It is ignored if we emit actual debug info. If we
   // don't, this at least helps the user find where a global came from.
   if (MAI->hasSingleParameterDotFile()) {
     // .file "foo.c"
+    if (MAI->hasBasenameOnlyForFileDirective())
       OutStreamer->emitFileDirective(
           llvm::sys::path::filename(M.getSourceFileName()));
+    else
+      OutStreamer->emitFileDirective(M.getSourceFileName());
   }
 
   GCModuleInfo *MI = getAnalysisIfAvailable<GCModuleInfo>();
   assert(MI && "AsmPrinter didn't require GCModuleInfo?");
   for (auto &I : *MI)
     if (GCMetadataPrinter *MP = GetOrCreateGCPrinter(*I))
       MP->beginAssembly(M, *MI, *this);
 

llvm/lib/MC/MCAsmInfoXCOFF.cpp

 
 extern cl::opt<cl::boolOrDefault> UseLEB128Directives;
 
 void MCAsmInfoXCOFF::anchor() {}
 
 MCAsmInfoXCOFF::MCAsmInfoXCOFF() {
   IsLittleEndian = false;
   HasVisibilityOnlyWithLinkage = true;
+  HasBasenameOnlyForFileDirective = false;
   PrivateGlobalPrefix = "L..";
   PrivateLabelPrefix = "L..";
   SupportsQuotedNames = false;
   UseDotAlignForAlignment = true;
   UsesDwarfFileAndLocDirectives = false;
   DwarfSectionSizeRequired = false;
   if (UseLEB128Directives == cl::BOU_UNSET)
     HasLEB128Directives = false;

llvm/test/CodeGen/PowerPC/aix-filename.ll

+; RUN: llc -verify-machineinstrs -mtriple powerpc-ibm-aix-xcoff < %s \
+; RUN:   | FileCheck %s
+; RUN: llc -verify-machineinstrs -mtriple powerpc64-ibm-aix-xcoff < %s \
+; RUN:   | FileCheck %s
+
+; CHECK: .file "/path/to/file"
+
+source_filename = "/path/to/file"