A set of small optimizations to reduce compilation time

Change-Id: I53292f03a19a928e49e3141082e53d34f070e6e8

Author: tbujewsk

visa/BinaryEncoding.cpp

@@ -2953,7 +2953,7 @@ inline BinaryEncoding::Status BinaryEncoding::ProduceBinaryInstructions()
     SetInstCounts((uint32_t)globalHalfInstNum);
 
     EncodingHelper::dumpOptReport(globalInstNum, numCompactedInst, numCompacted3SrcInst, kernel);
-    for (auto x = offsetVector.begin(); x != offsetVector.end(); x++)
+    for (auto x = offsetVector.begin(), vEnd = offsetVector.end(); x != vEnd; x++)
     {
         // calculate offsets again since labels for forward jumps/calls
         // are available now

visa/BinaryEncodingCNL.cpp

@@ -2578,7 +2578,7 @@ void BinaryEncodingCNL::DoAll()
 	SetInstCounts((uint32_t)globalHalfInstNum);
 
     EncodingHelper::dumpOptReport(globalInstNum, numCompactedInst, numCompacted3SrcInst, kernel);
-    for (auto x = offsetVector.begin(); x != offsetVector.end(); x++)
+    for (auto x = offsetVector.begin(), vEnd = offsetVector.end(); x != vEnd; x++)
     {
         if (!EncodeConditionalBranches(x->inst, x->offset))
         {

visa/BuildCISAIRImpl.cpp

@@ -443,12 +443,8 @@ void saveFCallState(G4_Kernel* kernel, savedFCallStates& savedFCallState)
     // the IR can be reused for another kernel rather than
     // recompiling.
     // kernel points to a stackcall function.
-    for( BB_LIST_ITER bb_it = kernel->fg.BBs.begin();
-        bb_it != kernel->fg.BBs.end();
-        bb_it++ )
+    for (auto curBB : kernel->fg.BBs)
     {
-        G4_BB* curBB = (*bb_it);
-
         if( curBB->size() > 0 && curBB->isEndWithFCall() )
         {
             // Save state for this fcall
@@ -542,8 +538,8 @@ void restoreFCallState(G4_Kernel* kernel, savedFCallStates& savedFCallState)
 
 G4_Kernel* Get_Resolved_Compilation_Unit( common_isa_header header, std::list<G4_Kernel*> compilation_units, int idx )
 {
-    for( std::list<G4_Kernel*>::iterator k = compilation_units.begin();
-        k != compilation_units.end(); k++ )
+    for( std::list<G4_Kernel*>::iterator k = compilation_units.begin(), c_end = compilation_units.end();
+        k != c_end; k++ )
     {
         if( (*k)->fg.builder->getCUnitId() == (header.num_kernels + idx) && (*k)->fg.builder->getIsKernel() == false )
         {
@@ -696,7 +692,7 @@ void Stitch_Compiled_Units( common_isa_header header, std::list<G4_Kernel*>& com
     }
 
     // Append declarations and color attributes from all callees to kernel
-    for (auto it = callee_index.begin(); it != callee_index.end(); ++it )
+    for (auto it = callee_index.begin(), ciEnd = callee_index.end(); it != ciEnd; ++it)
     {
         G4_Kernel* callee = Get_Resolved_Compilation_Unit( header, compilation_units, (*it) );
 
@@ -839,17 +835,17 @@ int CISA_IR_Builder::Compile( const char* nameInput)
 
         savedFCallStates savedFCallState;
 
-        for(std::list<VISAKernelImpl*>::iterator kernel_it = kernels.begin();
-            kernel_it != kernels.end();
+        for(std::list<VISAKernelImpl*>::iterator kernel_it = kernels.begin(), kend = kernels.end();
+            kernel_it != kend;
             kernel_it++)
         {
             VISAKernelImpl* kernel = (*kernel_it);
 
             saveFCallState(kernel->getKernel(), savedFCallState);
         }
 
-        for( std::list<VISAKernelImpl*>::iterator func_it = functions.begin();
-            func_it != functions.end();
+        for( std::list<VISAKernelImpl*>::iterator func_it = functions.begin(), fend = functions.end();
+            func_it != fend;
             func_it++ )
         {
             VISAKernelImpl* function = (*func_it);
@@ -1041,8 +1037,8 @@ int CISA_IR_Builder::CreateVISAFileVar(VISA_FileVar *& decl, char *varName, unsi
     if( IS_GEN_BOTH_PATH )
     {
         // Append file var to all kernel/function objects in CISA_IR_Builder
-        for( std::list<VISAKernelImpl*>::iterator it = m_kernels.begin();
-            it != m_kernels.end();
+        for( std::list<VISAKernelImpl*>::iterator it = m_kernels.begin(), kend = m_kernels.end();
+            it != kend;
             it++ )
         {
             VISAKernelImpl* kernel = (*it);

visa/CFGStructurizer.cpp

@@ -1159,7 +1159,7 @@ G4_INST* CFGStructurizer::getJmpiInst(G4_BB *bb)
 
 void CFGStructurizer::fini()
 {
-    for (uint32_t i = 0; i< topANodes.size(); ++i)
+    for (uint32_t i = 0, size = topANodes.size(); i< size; ++i)
     {
         deletePST((ANodeHG*)topANodes[i]);
     }
@@ -2169,7 +2169,7 @@ void CFGStructurizer::ANStack_reconstruct(ANodeHG *& node, ANodeBB *ndbb)
 
     // Merging from ANStackIx to the top of ANStack is needed to make
     // sure the order of children is correct.
-    for (uint32_t i = newNode->ANStackIx + 1; i < ANStack.size(); ++i)
+    for (uint32_t i = newNode->ANStackIx + 1, size = ANStack.size(); i < size; ++i)
     {
         ANodeHG *nd = (ANodeHG *)ANStack[i];
         getNewRange(end, exit, nd->getEndBB(), nd->getExitBB());
@@ -3656,7 +3656,7 @@ void CFGStructurizer::run()
     // Now, convert ANodes
     if (topANodes.size() > 0)
     {
-        for (uint32_t i = 0; i < topANodes.size(); ++i)
+        for (uint32_t i = 0, size = topANodes.size(); i < size; ++i)
         {
             (void)convertPST(topANodes[i], nullptr);
         }

visa/Common_BinaryEncoding.cpp

@@ -165,8 +165,8 @@ void BinaryEncodingBase::FixAlign16Inst(G4_INST* inst)
 	G4_DstRegRegion* dst = inst->getDst();
 	dst->setWriteMask(ChannelEnable_XYZW);
 
-	// convert sources to align16
-	for (int k = 0; k < inst->getNumSrc(); k++)
+	// convert sources to align16	
+    for (int k = 0, numSrc = inst->getNumSrc(); k < numSrc; k++)
 	{
 		ASSERT_USER(inst->getSrc(k)->isSrcRegRegion(), "Unexpected src to be converted to ALIGN16!");
 		G4_SrcRegRegion* src = inst->getSrc(k)->asSrcRegRegion();
@@ -232,7 +232,7 @@ void BinaryEncodingBase::FixAlign16Inst(G4_INST* inst)
 	// for double/half inst, we have to additionally fix the source as it doesn't support the .r swizzle
 	if (isDoubleInst)
 	{
-		for (int i = 0; i < inst->getNumSrc(); ++i)
+        for (int i = 0, numSrc = inst->getNumSrc(); i < numSrc; ++i)
 		{
 			MUST_BE_TRUE(inst->getSrc(i)->isSrcRegRegion(), "source must have a region");
 			G4_SrcRegRegion* src = inst->getSrc(i)->asSrcRegRegion();
@@ -254,7 +254,7 @@ void BinaryEncodingBase::FixAlign16Inst(G4_INST* inst)
 void BinaryEncodingBase::FixMathInst(G4_INST* inst)
 {
 	MUST_BE_TRUE(inst->isMath(), "Expect math instruction");
-	for (int i = 0; i < inst->getNumSrc(); ++i)
+    for (int i = 0, numSrc = inst->getNumSrc(); i < numSrc; ++i)
 	{
 		G4_Operand* src = inst->getSrc(i);
 		if (src && src->isSrcRegRegion())

visa/DebugInfo.cpp

@@ -547,7 +547,7 @@ void KernelDebugInfo::computeMissingVISAIds()
         }
     }
 
-    for (unsigned int i = 0; i < seenVISAIds.size(); i++)
+    for (unsigned int i = 0, size = seenVISAIds.size(); i < size; i++)
     {
         if (!seenVISAIds[i])
         {
@@ -627,7 +627,7 @@ void KernelDebugInfo::generateByteOffsetMapping(std::list<G4_BB*>& stackCallEntr
     unsigned int maxVISAIndex = 0;
     uint64_t maxGenIsaOffset = 0;
     // Now traverse CFG, create pair of CISA byte offset, gen binary offset and push to vector
-    for (BB_LIST_ITER bb_it = kernel->fg.BBs.begin(); bb_it != kernel->fg.BBs.end(); bb_it++)
+    for (BB_LIST_ITER bb_it = kernel->fg.BBs.begin(), bbEnd = kernel->fg.BBs.end(); bb_it != bbEnd; bb_it++)
     {
         G4_BB* bb = (*bb_it);
 
@@ -658,8 +658,8 @@ void KernelDebugInfo::generateByteOffsetMapping(std::list<G4_BB*>& stackCallEntr
             break;
         }
 
-        for(INST_LIST_ITER inst_it = bb->begin();
-            inst_it != bb->end();
+        for (INST_LIST_ITER inst_it = bb->begin(), bbEnd = bb->end();
+            inst_it != bbEnd;
             inst_it++)
         {
             G4_INST* inst = (*inst_it);
@@ -954,7 +954,7 @@ unsigned int populateMapDclName(VISAKernelImpl* kernel, std::map<G4_Declare*, st
 uint32_t KernelDebugInfo::getVarIndex(G4_Declare* dcl)
 {
     uint32_t retval = 0xffffffff;
-    for (uint32_t i = 0; i < varsMap.size(); i++)
+    for (uint32_t i = 0, size = varsMap.size(); i < size; i++)
     {
         if (dcl == varsMap[i]->dcl)
         {