[PATCH] MC: Split the x86 asm matcher implementations by dialect

[Reid Kleckner]

>>! In D4747#8, @grosbach wrote:
> I believe Daniel's right that WasOriginallyInvalidOperand is unused and can be nuked in MatchAndEmitIntelInstruction().

Gone.

> Do we already have testcases for the various other forms? I'm specifically thinking of the retry loop "for (unsigned Size : MopSizes) {".

It looks like I don't actually need the SSE sizes of 128, 256, and 512.  I can't find a single SSE instruction that lacks register operands, or doesn't encode the size into the mnemonic.

================
Comment at: test/MC/X86/intel-syntax-ambiguous.s:11
@@ +10,3 @@
+inc word ptr [eax]
+inc byte ptr [eax]
+// CHECK-NOT: error:
----------------
Jim Grosbach wrote:
> We should check the encoding of these instructions, not just that we can parse them without error.
I can't check encoding in the same test file.  Do you want a full encoding test, or is a simple gas-dump style test like intel-syntax.s good enough?

http://reviews.llvm.org/D4747