Add flag to allow thread combining for shader without barrier or SLM

Change-Id: Icd22324477ad59f9743865292e0d1c33a473b708

Author: Thomas

IGC/Compiler/ThreadCombining.cpp

@@ -355,7 +355,7 @@ bool ThreadCombining::canDoOptimization(Function* m_kernel, llvm::Module& M)
     if (threadGroupSize_X == 1 ||
 		threadGroupSize_Y == 1 ||
 		threadGroupSize_Z != 1 ||
-        !m_SLMUsed || 
+        (!m_SLMUsed && IGC_IS_FLAG_DISABLED(EnableThreadCombiningWithNoSLM))|| 
         anyBarrierWithinControlFlow)
     {
 		return false;

IGC/common/igc_flags.def

@@ -248,6 +248,7 @@ DECLARE_IGC_REGKEY(DWORD, OCLInlineThreshold,           512,   "Setting OCL inli
 DECLARE_IGC_REGKEY(bool, EnableForceGroupSize,          false, "Enable forcing thread Group Size ForceGroupSizeX and ForceGroupSizeY")
 DECLARE_IGC_REGKEY(DWORD, ForceGroupSizeX,              8, "force group size along X")
 DECLARE_IGC_REGKEY(DWORD, ForceGroupSizeY,              8, "force group size along Y")
+DECLARE_IGC_REGKEY(bool, EnableThreadCombiningWithNoSLM, false, "Enable thread combining opt for shader without SLM")
 DECLARE_IGC_REGKEY(DWORD, SubroutineThreshold,          110000, "Minimal kernel size to enable subroutines")
 DECLARE_IGC_REGKEY(DWORD, SubroutineInlinerThreshold,   3000, "Subroutine inliner threshold")
 DECLARE_IGC_REGKEY(bool, EnableConstantPromotion,       true, "Enable global constant data to register promotion")