Disable memopt on retry for common kernels.

Change memopt windows size to 1 for common kernels helps to avoid functional issues.
This workaround should be removed once the issue is resolved.

Author: ppogotov

IGC/Compiler/CISACodeGen/MemOpt.cpp

@@ -82,6 +82,8 @@ namespace {
         bool AllowNegativeSymPtrsForLoad = false;
         bool AllowVector8LoadStore = false;
 
+        unsigned Limit = IGC_GET_FLAG_VALUE(MemOptWindowSize);
+
         // Map of profit vector lengths per scalar type. Each entry specifies the
         // profit vector length of a given scalar type.
         // NOTE: Prepare the profit vector lengths in the *DESCENDING* order.
@@ -520,6 +522,18 @@ bool MemOpt::runOnFunction(Function& F) {
     CGC = getAnalysis<CodeGenContextWrapper>().getCodeGenContext();
     TLI = &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI();
 
+    auto isKernelWithForcedRetry = [this](Function* F) -> bool {
+        auto it = std::find(CGC->m_kernelsWithForcedRetry.begin(), CGC->m_kernelsWithForcedRetry.end(), F);
+        return (it != CGC->m_kernelsWithForcedRetry.end()) && !CGC->m_retryManager.IsFirstTry();
+    };
+    bool shouldMemOptWindowSizeBeReduced = isKernelWithForcedRetry(&F);
+
+    if (shouldMemOptWindowSizeBeReduced) {
+        Limit = 1;
+    } else {
+        Limit = IGC_GET_FLAG_VALUE(MemOptWindowSize);
+    }
+
     if (ProfitVectorLengths.empty())
         buildProfitVectorLengths(F);
 
@@ -636,7 +650,6 @@ bool MemOpt::removeRedBlockRead(GenIntrinsicInst* LeadingBlockRead,
     TrivialMemRefListTy& ToOpt, unsigned& sg_size)
 {
     MemRefListTy::iterator MI = aMI;
-    const unsigned Limit = IGC_GET_FLAG_VALUE(MemOptWindowSize);
     const unsigned windowEnd = Limit + MI->second;
     auto ME = MemRefs.end();
 
@@ -1208,7 +1221,6 @@ bool MemOpt::mergeLoad(LoadInst* LeadingLoad,
     // List of instructions need dependency check.
     SmallVector<Instruction*, 8> CheckList;
 
-    const unsigned Limit = IGC_GET_FLAG_VALUE(MemOptWindowSize);
     // Given the Start position of the Window is MI->second,
     // the End postion of the Window is "limit + Windows' start".
     const unsigned windowEnd = Limit + MI->second;
@@ -1562,7 +1574,6 @@ bool MemOpt::mergeStore(StoreInst* LeadingStore,
     // List of instructions need dependency check.
     SmallVector<Instruction*, 8> CheckList;
 
-    const unsigned Limit = IGC_GET_FLAG_VALUE(MemOptWindowSize);
     // Given the Start position of the Window is MI->second,
     // the End postion of the Window is "limit + Windows' start".
     const unsigned windowEnd = Limit + MI->second;

IGC/Compiler/CISACodeGen/ShaderCodeGen.cpp

@@ -366,8 +366,6 @@ void AddAnalysisPasses(CodeGenContext& ctx, IGCPassManager& mpm)
 
     mpm.add(new DpasScan());
 
-    mpm.add(new MatchCommonKernelPatterns());
-
     // let CleanPHINode be right before Layout
     mpm.add(createCleanPHINodePass());
     if(IGC_IS_FLAG_SET(DumpRegPressureEstimate)) mpm.add(new IGCRegisterPressurePrinter("final"));
@@ -723,6 +721,7 @@ void AddLegalizationPasses(CodeGenContext& ctx, IGCPassManager& mpm, PSSignature
     }
     // Should help MemOpt pass to merge more loads
     mpm.add(createSinkCommonOffsetFromGEPPass());
+    mpm.add(new MatchCommonKernelPatterns());
 
     // Run MemOpt
     if (!isOptDisabled &&