[XCOFF] Write source language ID and CPU version ID into C_FILE symbol.

[Esme]

Summary

The source language ID and CPU version ID are required by debuggers.
AIX's system assembler determines the source language ID based on the source file's name suffix, and the behavior in this patch is consistent with it.

[shchenz]

llvm/lib/MC/XCOFFObjectWriter.cpp

1116

This is not unreliable, for example you can compile files with many kinds of suffix or even without suffix as a c++ file if you specify -x c++. However seems we don't have accurate info in IR to identify the language ID if no -g. So I think it is OK here to have same handling with system assembler.

System assembler treats following suffix as C++ file:

.C, .cc, .ii, .cpp, .cxx, .c++
.ccm, .iim, .cppm, .cxxm, .c++m
.CC, .CPP, .CXX, .C++

Copied from @stephenpeckham

1123

Why set the default to 0XC?(assembly file? we won't use integrated-assembler to compile assembly files). Maybe we can default to C++ like traceback table does, and by doing this, we also have consistent language ID in the symbol table and the trace back table.

[shchenz]

llvm/lib/MC/XCOFFObjectWriter.cpp

1121

Please also handle the FORTRAN suffixs

[stephenpeckham]

llvm/lib/MC/XCOFFObjectWriter.cpp

1107–1108

You could use FileName instead of F.first.

By the way, the system assembler generates a C_FILE symbol with auxiliary symbol table entries, allowing the compiler version and timestamp to be specified. (That might need a separate patch.)

[stephenpeckham]

llvm/lib/MC/XCOFFObjectWriter.cpp

1120

Shouldn't this be "endswith" instead of "ends_with_insensitive"? Otherwise, .C looks like TB_C. Maybe you should check for .c first, then default to TB_CPLUSPLUS. (If you want checks for other languages, such as Fortran, you could add them.).

[Esme]

llvm/lib/MC/XCOFFObjectWriter.cpp

1121

Thanks!
However I'm not quite sure what suffixes the system's assembler would judge as FORTRAN?

[shchenz]

llvm/lib/MC/XCOFFObjectWriter.cpp

1121

Check here: https://www.ibm.com/docs/en/openxl-fortran-aix/17.1.0?topic=fortran-open-xl-input-output-files

[Esme]

Addressed comments. Thanks! @stephenpeckham @shchenz

[stephenpeckham]

I don't see a testcase for source_filename ending with ".c".

[hubert.reinterpretcast]

@Esme, iiuc, the integrated-as mode was made the default on AIX during the LLVM 17 development cycle and, without this, there is a regression in debugging Clang-compiled programs with AIX dbx?
If so, I believe this should be considered for the LLVM 17 release branch (i.e., backport if this does not land before the release split).

[Esme]

In D155684#4520202, @stephenpeckham wrote:

I don't see a testcase for source_filename ending with ".c".

Please see llvm/test/CodeGen/PowerPC/aix-filename-c.ll. Does it meet your expectation?

[Esme]

In D155684#4521185, @hubert.reinterpretcast wrote:

@Esme, iiuc, the integrated-as mode was made the default on AIX during the LLVM 17 development cycle and, without this, there is a regression in debugging Clang-compiled programs with AIX dbx?
If so, I believe this should be considered for the LLVM 17 release branch (i.e., backport if this does not land before the release split).

Yes, dbx requires the correct source language ID to locate symbols.
Thank you for reminding!

[stephenpeckham]

Please see llvm/test/CodeGen/PowerPC/aix-filename-c.ll. Does it meet your expectation?

Yes, sorry. I missed seeing that test earlier.

[shchenz]

LGTM. Thanks for fixing this.

llvm/include/llvm/BinaryFormat/XCOFF.h

enum CFileStringType : uint8_t {
   XFT_FN = 0,  ///< Specifies the source-file name.
   XFT_CT = 1,  ///< Specifies the compiler time stamp.
   XFT_CV = 2,  ///< Specifies the compiler version number.
   XFT_CD = 128 ///< Specifies compiler-defined information.
 };
 
 enum CFileLangId : uint8_t {
   TB_C = 0,        ///< C language.
+  TB_Fortran = 1,  ///< Fortran language.
   TB_CPLUSPLUS = 9 ///< C++ language.
 };
 
 enum CFileCpuId : uint8_t {
   TCPU_PPC64 = 2, ///< PowerPC common architecture 64-bit mode.
   TCPU_COM = 3,   ///< POWER and PowerPC architecture common.
   TCPU_970 = 19   ///< PPC970 - PowerPC 64-bit architecture.
 };

llvm/lib/MC/XCOFFObjectWriter.cpp

void XCOFFObjectWriter::writeRelocations() {
 
   for (const auto &DwarfSection : DwarfSections)
     for (const auto &Reloc : DwarfSection.DwarfSect->Relocations)
       writeRelocation(Reloc, *DwarfSection.DwarfSect);
 }
 
 void XCOFFObjectWriter::writeSymbolTable(const MCAsmLayout &Layout) {
   // Write C_FILE symbols.
+  for (const std::pair<std::string, size_t> &F : FileNames) {
     // The n_name of a C_FILE symbol is the source file's name when no auxiliary
     // entries are present.
-  for (const std::pair<std::string, size_t> &F : FileNames) {
-    writeSymbolEntry(F.first, /*Value=*/0, XCOFF::ReservedSectionNum::N_DEBUG,
-                     /*SymbolType=*/0, XCOFF::C_FILE,
+    StringRef FileName = F.first;
+
+    // For C_FILE symbols, the Source Language ID overlays the high-order byte
+    // of the SymbolType field, and the CPU Version ID is defined as the
+    // low-order byte.
+    // AIX's system assembler determines the source language ID based on the
+    // source file's name suffix, and the behavior here is consistent with it.
+    uint8_t LangID;
+    if (FileName.ends_with(".c"))
+      LangID = XCOFF::TB_C;
+    else if (FileName.ends_with_insensitive(".f") ||
+             FileName.ends_with_insensitive(".f77") ||
+             FileName.ends_with_insensitive(".f90") ||
+             FileName.ends_with_insensitive(".f95") ||
+             FileName.ends_with_insensitive(".f03") ||
+             FileName.ends_with_insensitive(".f08"))
+      LangID = XCOFF::TB_Fortran;
+    else
+      LangID = XCOFF::TB_CPLUSPLUS;
+    uint8_t CpuID;
+    if (is64Bit())
+      CpuID = XCOFF::TCPU_PPC64;
+    else
+      CpuID = XCOFF::TCPU_COM;
+
+    writeSymbolEntry(FileName, /*Value=*/0, XCOFF::ReservedSectionNum::N_DEBUG,
+                     /*SymbolType=*/(LangID << 8) | CpuID, XCOFF::C_FILE,
                      /*NumberOfAuxEntries=*/0);
   }
 
   if (CInfoSymSection.Entry)
     writeSymbolEntry(CInfoSymSection.Entry->Name, CInfoSymSection.Entry->Offset,
                      CInfoSymSection.Index,
                      /*SymbolType=*/0, XCOFF::C_INFO,
                      /*NumberOfAuxEntries=*/0);

llvm/test/CodeGen/PowerPC/aix-extern-weak.ll

 ; COMMON-NEXT:    .tc b_w[TC],b_w[UA]
 
 ; CHECKSYM:      Symbols [
 ; CHECKSYM-NEXT:   Symbol {
 ; CHECKSYM-NEXT:     Index: 0
 ; CHECKSYM-NEXT:     Name: <stdin>
 ; CHECKSYM-NEXT:     Value (SymbolTableIndex): 0x0
 ; CHECKSYM-NEXT:     Section: N_DEBUG
-; CHECKSYM-NEXT:     Source Language ID: TB_C (0x0)
-; CHECKSYM-NEXT:     CPU Version ID: 0x0
+; CHECKSYM-NEXT:     Source Language ID: TB_CPLUSPLUS (0x9)
+; CHECKSYM32-NEXT:   CPU Version ID: TCPU_COM (0x3)
+; CHECKSYM64-NEXT:   CPU Version ID: TCPU_PPC64 (0x2)
 ; CHECKSYM-NEXT:     StorageClass: C_FILE (0x67)
 ; CHECKSYM-NEXT:     NumberOfAuxEntries: 0
 ; CHECKSYM-NEXT:   }
 ; CHECKSYM-NEXT:   Symbol {
 ; CHECKSYM-NEXT:     Index: [[#Index:]]
 ; CHECKSYM-NEXT:     Name: .foo_ext_weak
 ; CHECKSYM-NEXT:     Value (RelocatableAddress): 0x0
 ; CHECKSYM-NEXT:     Section: N_UNDEF

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

 ; COMMON-NEXT:      .tc bar_p[TC],bar_p
 
 ; CHECKSYM:       Symbols [
 ; CHECKSYM-NEXT:   Symbol {
 ; CHECKSYM-NEXT:     Index: 0
 ; CHECKSYM-NEXT:     Name: <stdin>
 ; CHECKSYM-NEXT:     Value (SymbolTableIndex): 0x0
 ; CHECKSYM-NEXT:     Section: N_DEBUG
-; CHECKSYM-NEXT:     Source Language ID: TB_C (0x0)
-; CHECKSYM-NEXT:     CPU Version ID: 0x0
+; CHECKSYM-NEXT:     Source Language ID: TB_CPLUSPLUS (0x9)
+; CHECKSYM32-NEXT:   CPU Version ID: TCPU_COM (0x3)
+; CHECKSYM64-NEXT:   CPU Version ID: TCPU_PPC64 (0x2)
 ; CHECKSYM-NEXT:     StorageClass: C_FILE (0x67)
 ; CHECKSYM-NEXT:     NumberOfAuxEntries: 0
 ; CHECKSYM-NEXT:   }
 ; CHECKSYM-NEXT:   Symbol {
 ; CHECKSYM-NEXT:     Index: [[#Index:]]
 ; CHECKSYM-NEXT:     Name: .bar_extern
 ; CHECKSYM-NEXT:     Value (RelocatableAddress): 0x0
 ; CHECKSYM-NEXT:     Section: N_UNDEF

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

+; RUN: llc -verify-machineinstrs -mtriple powerpc-ibm-aix-xcoff -filetype=obj -o %t.o < %s
+; RUN: llvm-readobj --symbols %t.o | FileCheck --check-prefixes=OBJ,OBJ32 %s
+; RUN: llc -verify-machineinstrs -mtriple powerpc64-ibm-aix-xcoff -filetype=obj -o %t64.o < %s
+; RUN: llvm-readobj --symbols %t64.o | FileCheck --check-prefixes=OBJ,OBJ64 %s
+
+source_filename = "1.c"
+
+; OBJ: Name: 1.c
+; OBJ: Source Language ID: TB_C (0x0)
+; OBJ32: CPU Version ID: TCPU_COM (0x3)
+; OBJ64: CPU Version ID: TCPU_PPC64 (0x2)

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

+; RUN: llc -verify-machineinstrs -mtriple powerpc-ibm-aix-xcoff -filetype=obj -o %t.o < %s
+; RUN: llvm-readobj --symbols %t.o | FileCheck --check-prefixes=OBJ,OBJ32 %s
+; RUN: llc -verify-machineinstrs -mtriple powerpc64-ibm-aix-xcoff -filetype=obj -o %t64.o < %s
+; RUN: llvm-readobj --symbols %t64.o | FileCheck --check-prefixes=OBJ,OBJ64 %s
+
+source_filename = "1.cpp"
+
+; OBJ: Name: 1.cpp
+; OBJ: Source Language ID: TB_CPLUSPLUS (0x9)
+; OBJ32: CPU Version ID: TCPU_COM (0x3)
+; OBJ64: CPU Version ID: TCPU_PPC64 (0x2)

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

+; RUN: llc -verify-machineinstrs -mtriple powerpc-ibm-aix-xcoff -filetype=obj -o %t.o < %s
+; RUN: llvm-readobj --symbols %t.o | FileCheck --check-prefixes=OBJ,OBJ32 %s
+; RUN: llc -verify-machineinstrs -mtriple powerpc64-ibm-aix-xcoff -filetype=obj -o %t64.o < %s
+; RUN: llvm-readobj --symbols %t64.o | FileCheck --check-prefixes=OBJ,OBJ64 %s
+
+source_filename = "1.f95"
+
+; OBJ: Name: 1.f95
+; OBJ: Source Language ID: TB_Fortran (0x1)
+; OBJ32: CPU Version ID: TCPU_COM (0x3)
+; OBJ64: CPU Version ID: TCPU_PPC64 (0x2)

llvm/test/CodeGen/PowerPC/aix-func-dsc-gen.ll

 ; CHECK-NEXT: Format: aixcoff-rs6000
 ; CHECK-NEXT: Arch: powerpc
 ; CHECK-NEXT: AddressSize: 32bit
 ; CHECK:        Symbol {
 ; CHECK-NEXT:     Index: 0
 ; CHECK-NEXT:     Name: <stdin>
 ; CHECK-NEXT:     Value (SymbolTableIndex): 0x0
 ; CHECK-NEXT:     Section: N_DEBUG
-; CHECK-NEXT:     Source Language ID: TB_C (0x0)
-; CHECK-NEXT:     CPU Version ID: 0x0
+; CHECK-NEXT:     Source Language ID: TB_CPLUSPLUS (0x9)
+; CHECK-NEXT:     CPU Version ID: TCPU_COM (0x3)
 ; CHECK-NEXT:     StorageClass: C_FILE (0x67)
 ; CHECK-NEXT:     NumberOfAuxEntries: 0
 ; CHECK-NEXT:   }
 ; CHECK-NEXT:   Symbol {
 ; CHECK-NEXT:     Index: [[#Index:]]
 ; CHECK-NEXT:     Name:
 ; CHECK-NEXT:     Value (RelocatableAddress): 0x0
 ; CHECK-NEXT:     Section: .text

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

 ; CHECK64:              bl .___memset64
 ; CHECK64:              .extern .___memset64
 
 ; CHECKSYM:        Symbol {
 ; CHECKSYM-NEXT:     Index: 0
 ; CHECKSYM-NEXT:     Name: <stdin>
 ; CHECKSYM-NEXT:     Value (SymbolTableIndex): 0x0
 ; CHECKSYM-NEXT:     Section: N_DEBUG
-; CHECKSYM-NEXT:     Source Language ID: TB_C (0x0)
-; CHECKSYM-NEXT:     CPU Version ID: 0x0
+; CHECKSYM-NEXT:     Source Language ID: TB_CPLUSPLUS (0x9)
+; CHECKSYM32-NEXT:   CPU Version ID: TCPU_COM (0x3)
+; CHECKSYM64-NEXT:   CPU Version ID: TCPU_PPC64 (0x2)
 ; CHECKSYM-NEXT:     StorageClass: C_FILE (0x67)
 ; CHECKSYM-NEXT:     NumberOfAuxEntries: 0
 ; CHECKSYM-NEXT:   }
 ; CHECKSYM-NEXT:   Symbol {
 ; CHECKSYM-NEXT:     Index: 1
 ; CHECKSYM32-NEXT:     Name: .___memset
 ; CHECKSYM64-NEXT:     Name: .___memset64
 ; CHECKSYM-NEXT:     Value (RelocatableAddress): 0x0

llvm/test/CodeGen/PowerPC/aix-tls-xcoff-reloc-large.ll

 ; SYM-NEXT: Arch: powerpc
 ; SYM-NEXT: AddressSize: 32bit
 ; SYM-NEXT: Symbols [
 ; SYM-NEXT:   Symbol {
 ; SYM-NEXT:     Index: 0
 ; SYM-NEXT:     Name: <stdin>
 ; SYM-NEXT:     Value (SymbolTableIndex): 0x0
 ; SYM-NEXT:     Section: N_DEBUG
-; SYM-NEXT:     Source Language ID: TB_C (0x0)
-; SYM-NEXT:     CPU Version ID: 0x0
+; SYM-NEXT:     Source Language ID: TB_CPLUSPLUS (0x9)
+; SYM-NEXT:     CPU Version ID: TCPU_COM (0x3)
 ; SYM-NEXT:     StorageClass: C_FILE (0x67)
 ; SYM-NEXT:     NumberOfAuxEntries: 0
 ; SYM-NEXT:   }
 ; SYM-NEXT:   Symbol {
 ; SYM-NEXT:     Index: 1
 ; SYM-NEXT:     Name: .__tls_get_addr
 ; SYM-NEXT:     Value (RelocatableAddress): 0x0
 ; SYM-NEXT:     Section: N_UNDEF

llvm/test/CodeGen/PowerPC/aix-tls-xcoff-reloc.ll

 ; SYM-NEXT: Arch: powerpc
 ; SYM-NEXT: AddressSize: 32bit
 ; SYM-NEXT: Symbols [
 ; SYM-NEXT:   Symbol {
 ; SYM-NEXT:     Index: 0
 ; SYM-NEXT:     Name: <stdin>
 ; SYM-NEXT:     Value (SymbolTableIndex): 0x0
 ; SYM-NEXT:     Section: N_DEBUG
-; SYM-NEXT:     Source Language ID: TB_C (0x0)
-; SYM-NEXT:     CPU Version ID: 0x0
+; SYM-NEXT:     Source Language ID: TB_CPLUSPLUS (0x9)
+; SYM-NEXT:     CPU Version ID: TCPU_COM (0x3)
 ; SYM-NEXT:     StorageClass: C_FILE (0x67)
 ; SYM-NEXT:     NumberOfAuxEntries: 0
 ; SYM-NEXT:   }
 ; SYM-NEXT:   Symbol {
 ; SYM-NEXT:     Index: 1
 ; SYM-NEXT:     Name: .__tls_get_addr
 ; SYM-NEXT:     Value (RelocatableAddress): 0x0
 ; SYM-NEXT:     Section: N_UNDEF

llvm/test/CodeGen/PowerPC/aix-tls-xcoff-variables.ll

 ; SYMS-NEXT: Arch: powerpc
 ; SYMS-NEXT: AddressSize: 32bit
 ; SYMS-NEXT: Symbols [
 ; SYMS-NEXT:   Symbol {
 ; SYMS-NEXT:     Index: 0
 ; SYMS-NEXT:     Name: <stdin>
 ; SYMS-NEXT:     Value (SymbolTableIndex): 0x0
 ; SYMS-NEXT:     Section: N_DEBUG
-; SYMS-NEXT:     Source Language ID: TB_C (0x0)
-; SYMS-NEXT:     CPU Version ID: 0x0
+; SYMS-NEXT:     Source Language ID: TB_CPLUSPLUS (0x9)
+; SYMS-NEXT:     CPU Version ID: TCPU_COM (0x3)
 ; SYMS-NEXT:     StorageClass: C_FILE (0x67)
 ; SYMS-NEXT:     NumberOfAuxEntries: 0
 ; SYMS-NEXT:   }
 ; SYMS-NEXT:   Symbol {
 ; SYMS-NEXT:     Index: [[#INDX:]]
 ; SYMS-NEXT:     Name: tls_global_int_external_uninitialized
 ; SYMS-NEXT:     Value (RelocatableAddress): 0x0
 ; SYMS-NEXT:     Section: N_UNDEF

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

 
 
 ; CHECKSYM:      Symbols [
 ; CHECKSYM-NEXT:   Symbol {
 ; CHECKSYM-NEXT:     Index: 0
 ; CHECKSYM-NEXT:     Name: <stdin>
 ; CHECKSYM-NEXT:     Value (SymbolTableIndex): 0x0
 ; CHECKSYM-NEXT:     Section: N_DEBUG
-; CHECKSYM-NEXT:     Source Language ID: TB_C (0x0)
-; CHECKSYM-NEXT:     CPU Version ID: 0x0
+; CHECKSYM-NEXT:     Source Language ID: TB_CPLUSPLUS (0x9)
+; CHECKSYM32-NEXT:   CPU Version ID: TCPU_COM (0x3)
+; CHECKSYM64-NEXT:   CPU Version ID: TCPU_PPC64 (0x2)
 ; CHECKSYM-NEXT:     StorageClass: C_FILE (0x67)
 ; CHECKSYM-NEXT:     NumberOfAuxEntries: 0
 ; CHECKSYM-NEXT:   }
 ; CHECKSYM-NEXT:   Symbol {
 ; CHECKSYM-NEXT:     Index: [[#Index:]]
 ; CHECKSYM-NEXT:     Name: 
 ; CHECKSYM-NEXT:     Value (RelocatableAddress): 0x0
 ; CHECKSYM-NEXT:     Section: .text

llvm/test/CodeGen/PowerPC/aix-xcoff-data.ll

 ; OBJ:      ]
 
 ; SYMS:      Symbols [
 ; SYMS-NEXT:   Symbol {
 ; SYMS-NEXT:     Index: 0
 ; SYMS-NEXT:     Name: <stdin>
 ; SYMS-NEXT:     Value (SymbolTableIndex): 0x0
 ; SYMS-NEXT:     Section: N_DEBUG
-; SYMS-NEXT:     Source Language ID: TB_C (0x0)
-; SYMS-NEXT:     CPU Version ID: 0x0
+; SYMS-NEXT:     Source Language ID: TB_CPLUSPLUS (0x9)
+; SYMS32-NEXT:   CPU Version ID: TCPU_COM (0x3)
+; SYMS64-NEXT:   CPU Version ID: TCPU_PPC64 (0x2)
 ; SYMS-NEXT:     StorageClass: C_FILE (0x67)
 ; SYMS-NEXT:     NumberOfAuxEntries: 0
 ; SYMS-NEXT:   }
 ; SYMS-NEXT:   Symbol {
 ; SYMS-NEXT:     Index: [[#INDX:]]
 ; SYMS-NEXT:     Name:
 ; SYMS-NEXT:     Value (RelocatableAddress): 0x0
 ; SYMS-NEXT:     Section: .text

llvm/test/CodeGen/PowerPC/aix-xcoff-reloc.ll

 ; RELOC-NEXT: ]
 
 ; SYM:      Symbols [
 ; SYM-NEXT:   Symbol {
 ; SYM-NEXT:     Index: 0
 ; SYM-NEXT:     Name: <stdin>
 ; SYM-NEXT:     Value (SymbolTableIndex): 0x0
 ; SYM-NEXT:     Section: N_DEBUG
-; SYM-NEXT:     Source Language ID: TB_C (0x0)
-; SYM-NEXT:     CPU Version ID: 0x0
+; SYM-NEXT:     Source Language ID: TB_CPLUSPLUS (0x9)
+; SYM32-NEXT:   CPU Version ID: TCPU_COM (0x3)
+; SYM64-NEXT:   CPU Version ID: TCPU_PPC64 (0x2)
 ; SYM-NEXT:     StorageClass: C_FILE (0x67)
 ; SYM-NEXT:     NumberOfAuxEntries: 0
 ; SYM-NEXT:   }
 ; SYM-NEXT:   Symbol {
 ; SYM-NEXT:     Index: [[#INDX:]]
 ; SYM-NEXT:     Name: .bar
 ; SYM-NEXT:     Value (RelocatableAddress): 0x0
 ; SYM-NEXT:     Section: N_UNDEF

llvm/tools/llvm-readobj/XCOFFDumper.cpp

static StringRef GetSymbolValueName(XCOFF::StorageClass SC) {
   default:
     return "Value";
   }
 }
 
 const EnumEntry<XCOFF::CFileLangId> CFileLangIdClass[] = {
 #define ECase(X)                                                               \
   { #X, XCOFF::X }
-    ECase(TB_C), ECase(TB_CPLUSPLUS)
+    ECase(TB_C), ECase(TB_Fortran), ECase(TB_CPLUSPLUS)
 #undef ECase
 };
 
 const EnumEntry<XCOFF::CFileCpuId> CFileCpuIdClass[] = {
 #define ECase(X)                                                               \
   { #X, XCOFF::X }
     ECase(TCPU_PPC64), ECase(TCPU_COM), ECase(TCPU_970)
 #undef ECase