[llvm-objdump] Default to --mattr=+all for AArch64

[MaskRay]

Summary

GNU objdump disassembles all unknown instructions by default. Match this user
friendly behavior with the target feature "all" (D128029) designed for disassemblers.

[jhenderson]

Seems good to me, but might want to give a chance for AArch64 contributors to have a say in the matter.

[MaskRay]

llvm/docs/CommandGuide/llvm-objdump.rst

32

typo: will change one to mcpu

[nickdesaulniers]

llvm/tools/llvm-objdump/llvm-objdump.cpp

1696–1698

Adding context to my comment, this is where I'd expect us to handle per-arch enabling of various target features.

...
if (MAttrs.empty() && MCPU.empty())
  enableAllArchFeatures()
...

static void enableAllArchFeatures() {
  switch (arch) {
  case aarch64:
    Features.AddFeature("+sve", etc etc);
    break;
  case x86:
...
}

[MaskRay]

llvm/tools/llvm-objdump/llvm-objdump.cpp

1696–1698

Resolved in D128029 ((a) Use TableGen to derive the source of truth. (b) The disassembler side just forwards "+all")

[MaskRay]

fix a typo. rebase

[MaskRay]

In case folks (@kongyi) want to cherry pick D128029 and D128030 into an old release branch. f4b7b66c4c75d40c099ef41ecea0d3be27bdb6c1 ([AArch64][test] Add --mattr=-{sve,sve2,sme} to SVE/SVE2/SME MC tests) is also needed
(it likely does not apply cleanly since there has been lots of SVE development).

llvm/docs/CommandGuide/llvm-objdump.rst

 combined with other commands:
 
 .. option:: -a, --archive-headers
 
   Display the information contained within an archive's headers.
 
 .. option:: -d, --disassemble
 
-  Disassemble all executable sections found in the input files.
+  Disassemble all executable sections found in the input files. On some
+  architectures (AArch64, x86), all known instructions are disassembled by
+  default. On the others, :option:`--mcpu` or :option:`--mattr` is needed to
+  enable some instruction sets. Disabled instructions are displayed as
+  ``<unknown>``.
 
 .. option:: -D, --disassemble-all
 
   Disassemble all sections found in the input files.
 
 .. option:: --disassemble-symbols=<symbol1[,symbol2,...]>
 
   Disassemble only the specified symbols. Takes demangled symbol names when

llvm/test/tools/llvm-objdump/ELF/AArch64/mattr.s

+## When --mattr and --mcpu are both empty, disassemble all known instructions.
+# RUN: llvm-mc -filetype=obj -triple=aarch64 -mattr=+all %s -o %t
+# RUN: llvm-objdump -d --no-show-raw-insn %t | FileCheck %s --check-prefixes=CHECK,ALL
+
+## If --mattr or --mcpu is specified, don't default to --mattr=+all.
+# RUN: llvm-objdump -d --no-show-raw-insn --mattr=+v8a %t | FileCheck %s --check-prefixes=CHECK,UNKNOWN
+# RUN: llvm-objdump -d --no-show-raw-insn --mcpu=generic %t | FileCheck %s --check-prefixes=CHECK,UNKNOWN
+
+# CHECK:    <_start>:
+# ALL-NEXT:   bc.eq 0x4
+# ALL-NEXT:   irg x0, x1
+# ALL-NEXT:   mrs x0, RNDR
+# UNKNOWN-COUNT-2: <unknown>
+# UNKNOWN:    mrs x0, S3_3_C2_C4_0
+# CHECK-EMPTY:
+
+.globl _start
+_start:
+  bc.eq #4      // armv8.8-a hbc
+  irg x0, x1    // armv8.5-a mte
+  mrs x0, RNDR  // armv8.5-a rand

llvm/tools/llvm-objdump/llvm-objdump.cpp

for (StringRef Sym : MissingDisasmSymbolSet.keys())
     reportWarning("failed to disassemble missing symbol " + Sym, FileName);
 }
 
 static void disassembleObject(const ObjectFile *Obj, bool InlineRelocs) {
   const Target *TheTarget = getTarget(Obj);
 
   // Package up features to be passed to target/subtarget
   SubtargetFeatures Features = Obj->getFeatures();
-  if (!MAttrs.empty())
+  if (!MAttrs.empty()) {
     for (unsigned I = 0; I != MAttrs.size(); ++I)
       Features.AddFeature(MAttrs[I]);
+  } else if (MCPU.empty() && Obj->getArch() == llvm::Triple::aarch64) {
+    Features.AddFeature("+all");
+  }
 
   std::unique_ptr<const MCRegisterInfo> MRI(
       TheTarget->createMCRegInfo(TripleName));
   if (!MRI)
     reportError(Obj->getFileName(),
                 "no register info for target " + TripleName);
 
   // Set up disassembler.