[AsmParser] Make generic directives and aliases case insensitive.

[DavidSpickett]

Summary

GCC will accept any case for assembler directives.
For example ".abort" and ".ABORT" (even ".aBoRt")
are equivalent.

https://sourceware.org/binutils/docs/as/Pseudo-Ops.html#Pseudo-Ops
"The names are case insensitive for most targets,
and usually written in lower case."

Change llvm-mc to accept any case for generic directives
or aliases of those directives.

This for Bugzilla #39527.

[DavidSpickett]

Direct link to Bugzilla: https://bugs.llvm.org/show_bug.cgi?id=39527

I chose to test one directive and one directive alias, instead of trying to create a file with one of every alias. Maybe it should have a few more than one of each though.

[ostannard]

"The names are case insensitive for most targets, and usually written in lower case."

This suggests that there are exceptions, do you which targets and directives don't follow this?

llvm/lib/MC/MCParser/AsmParser.cpp

5323

Why is this change needed, when all the strings are already lower-case?

[DavidSpickett]

llvm/lib/MC/MCParser/AsmParser.cpp

5323

An attempt to make it harder to put mixed case in, though you could just DirectiveKindMap[<bla>] directly anyway.

I'll leave initializeDirective as it is and just add a comment at the top of the function saying that they should be lower case.

[DavidSpickett]

This suggests that there are exceptions, do you which targets and directives don't follow this?

In general, only ABORT is called out as allowed for COFF output, which is more of a compatibility note. aBoRt works fine too.

I went through the AS target documentation and found the following targets where directives were not all lower case:

• ARC
• MMIX (all upper case, though GCC accepts any)
• V850

Along the way I realized that in LLVM each target has it's own <target>AsmParser::ParseDirective. Some of which use .lower() some don't.

So I think I'll reduce the scope of this to allow any case for the generic directives and aliases of those. (I don't see any alias that isn't all lower case)

That means that target specific directives are still case sensitive unless I go through each parser. Which I can do, just not sure whether it's the right way to go.

[DavidSpickett]

Added a comment to init directive map saying that they should all be lower case.
Switched to a generic x86 target and alias for the lit test, to match the other MC/AsmParser tests.

[ostannard]

LGTM

llvm/lib/MC/MCParser/AsmParser.cpp

public:
   bool Run(bool NoInitialTextSection, bool NoFinalize = false) override;
 
   void addDirectiveHandler(StringRef Directive,
                            ExtensionDirectiveHandler Handler) override {
     ExtensionDirectiveMap[Directive] = Handler;
   }
 
   void addAliasForDirective(StringRef Directive, StringRef Alias) override {
-    DirectiveKindMap[Directive] = DirectiveKindMap[Alias];
+    DirectiveKindMap[Directive.lower()] = DirectiveKindMap[Alias.lower()];
   }
 
   /// @name MCAsmParser Interface
   /// {
 
   SourceMgr &getSourceManager() override { return SrcMgr; }
   MCAsmLexer &getLexer() override { return Lexer; }
   MCContext &getContext() override { return Ctx; }
if (Lexer.is(AsmToken::Integer)) {
     }
     IDVal = "";
   }
 
   // Handle conditional assembly here before checking for skipping.  We
   // have to do this so that .endif isn't skipped in a ".if 0" block for
   // example.
   StringMap<DirectiveKind>::const_iterator DirKindIt =
-      DirectiveKindMap.find(IDVal);
+      DirectiveKindMap.find(IDVal.lower());
   DirectiveKind DirKind = (DirKindIt == DirectiveKindMap.end())
 
                               ? DK_NO_DIRECTIVE
                               : DirKindIt->getValue();
   switch (DirKind) {
   default:
     break;
   case DK_IF:
if (!TheCondStack.empty()) {
     TheCondState = TheCondStack.back();
     TheCondStack.pop_back();
   }
 
   return false;
 }
 
 void AsmParser::initializeDirectiveKindMap() {
+  /* Lookup will be done with the directive
+   * converted to lower case, so all these
+   * keys should be lower case.
+   * (target specific directives are handled
+   *  elsewhere)
+   */
   DirectiveKindMap[".set"] = DK_SET;
   DirectiveKindMap[".equ"] = DK_EQU;
   DirectiveKindMap[".equiv"] = DK_EQUIV;
   DirectiveKindMap[".ascii"] = DK_ASCII;
   DirectiveKindMap[".asciz"] = DK_ASCIZ;
   DirectiveKindMap[".string"] = DK_STRING;
   DirectiveKindMap[".byte"] = DK_BYTE;
   DirectiveKindMap[".short"] = DK_SHORT;

llvm/test/MC/AsmParser/directive_case_insensitive.s

+# RUN: llvm-mc -triple i386-unknown-unknown %s | FileCheck %s
+
+# CHECK: .byte 65
+        .ascii "A"
+# CHECK: .byte 66
+        .ASCII "B"
+# CHECK: .byte 67
+        .aScIi "C"
+
+# Note: using 2byte because it is an alias
+# CHECK: .short 4660
+        .2byte 0x1234
+# CHECK: .short 4661
+        .2BYTE 0x1235
+# CHECK: .short 4662
+        .2bYtE 0x1236