[extract_symbols.py] Filter out more symbols for MSVC

[glandium]

Summary

This strips out about 5k symbols.

Fixes https://github.com/llvm/llvm-project/issues/60109

[glandium]

llvm/utils/extract_symbols.py

171

Actually... they are, and they do :(

[glandium]

Switched to skipping X86GenMnemonicTables functions instead of private methods.

[john.brawn]

llvm/utils/extract_symbols.py

146

I think we can also discard __ymm symbols here (from looking at llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp).

171

Could you add something to the comment here saying why discarding these symbols is a valid thing to do?

[glandium]

Updated per feedback.

[john.brawn]

LGTM

llvm/utils/extract_symbols.py

 # required and which we can discard.
 def should_keep_microsoft_symbol(symbol, calling_convention_decoration):
     # Keep unmangled (i.e. extern "C") names
     if not '?' in symbol:
         if calling_convention_decoration:
             # Remove calling convention decoration from names
             match = re.match('[_@]([^@]+)', symbol)
             if match:
-                return match.group(1)
+                symbol = match.group(1)
+        # Discard floating point/SIMD constants.
+        if symbol.startswith(("__xmm@", "__ymm@", "__real@")):
+            return None
         return symbol
     # Function template instantiations start with ?$; keep the instantiations of
     # clang::Type::getAs, as some of them are explipict specializations that are
     # defined in clang's lib/AST/Type.cpp; discard the rest as it's assumed that
     # the definition is public
     elif re.match('\?\?\$getAs@.+@Type@clang@@', symbol):
         return symbol
     elif symbol.startswith('??$'):
         return None
     # Deleting destructors start with ?_G or ?_E and can be discarded because
     # link.exe gives you a warning telling you they can't be exported if you
     # don't
     elif symbol.startswith('??_G') or symbol.startswith('??_E'):
         return None
     # Constructors (?0) and destructors (?1) of templates (?$) are assumed to be
     # defined in headers and not required to be kept
     elif symbol.startswith('??0?$') or symbol.startswith('??1?$'):
         return None
     # An anonymous namespace is mangled as ?A(maybe hex number)@. Any symbol
     # that mentions an anonymous namespace can be discarded, as the anonymous
     # namespace doesn't exist outside of that translation unit.
     elif re.search('\?A(0x\w+)?@', symbol):
         return None
+    # Skip X86GenMnemonicTables functions, they are not exposed from llvm/include/.
+    elif re.match('\?is[A-Z0-9]*@X86@llvm', symbol):
+        return None
     # Keep mangled llvm:: and clang:: function symbols. How we detect these is a
     # bit of a mess and imprecise, but that avoids having to completely demangle
     # the symbol name. The outermost namespace is at the end of the identifier
     # mangling, and the identifier mangling is followed by the type mangling, so
     # we look for (llvm|clang)@@ followed by something that looks like a
     # function type mangling. To spot a function type we use (this is derived
     # from clang/lib/AST/MicrosoftMangle.cpp):
     # <function-type> ::= <function-class> <this-cvr-qualifiers>