AMDGPU: Fix maximum instruction size for amdgcn

[arsenm]

Summary

This was causing the conservative estimate of inline asm to be twice as big as expected.

[• tstellarAMD]

LGTM.

[arsenm]

r272956

lib/Target/AMDGPU/MCTargetDesc/AMDGPUMCAsmInfo.cpp

 //===-- MCTargetDesc/AMDGPUMCAsmInfo.cpp - Assembly Info ------------------===//
 //
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 /// \file
 //===----------------------------------------------------------------------===//
 
 #include "AMDGPUMCAsmInfo.h"
+#include "llvm/ADT/Triple.h"
 
 using namespace llvm;
+
 AMDGPUMCAsmInfo::AMDGPUMCAsmInfo(const Triple &TT) : MCAsmInfoELF() {
   HasSingleParameterDotFile = false;
   //===------------------------------------------------------------------===//
-  MaxInstLength = 16;
+  MaxInstLength = (TT.getArch() == Triple::amdgcn) ? 8 : 16;
   SeparatorString = "\n";
   CommentString = ";";
   PrivateLabelPrefix = "";
   InlineAsmStart = ";#ASMSTART";
   InlineAsmEnd = ";#ASMEND";
 
   //===--- Data Emission Directives -------------------------------------===//
   SunStyleELFSectionSwitchSyntax = true;