Revisit MismatchDetected for LowerGEPForPrivMem

MismatchDetected solved assert in the PrivateMemoryResolution pass
caused by mismatched widths. LowerGEPForPrivMem solves that case
but in case of alloca over the allowed size it will leave mismatched
widths for the later PrivateMemoryResolution causing the assert to
fail. Extended mismatch detection for cases with struct of array/struct
of vector.

Author: jszt1

IGC/Compiler/CISACodeGen/LowerGEPForPrivMem.cpp

@@ -282,7 +282,7 @@ StatusPrivArr2Reg LowerGEPForPrivMem::CheckIfAllocaPromotable(llvm::AllocaInst *
 
       allowedAllocaSizeInBytes = (allowedAllocaSizeInBytes * 8) / SIMDSize;
     }
-  SOALayoutChecker checker(*pAlloca, m_ctx->type == ShaderType::OPENCL_SHADER);
+  SOALayoutChecker checker(*pAlloca, m_ctx->type == ShaderType::OPENCL_SHADER, true);
   SOALayoutInfo SOAInfo = checker.getOrGatherInfo();
   if (!SOAInfo.canUseSOALayout) {
     return StatusPrivArr2Reg::CannotUseSOALayout;
@@ -357,7 +357,7 @@ StatusPrivArr2Reg LowerGEPForPrivMem::CheckIfAllocaPromotable(llvm::AllocaInst *
   return StatusPrivArr2Reg::OK;
 }
 
-SOALayoutChecker::SOALayoutChecker(AllocaInst &allocaToCheck, bool isOCL) : allocaRef(allocaToCheck) {
+SOALayoutChecker::SOALayoutChecker(AllocaInst &allocaToCheck, bool isOCL, bool mismatchedWidthsSupport) : allocaRef(allocaToCheck), mismatchedWidthsSupport(mismatchedWidthsSupport) {
   auto F = allocaToCheck.getParent()->getParent();
   pDL = &F->getParent()->getDataLayout();
   newAlgoControl = IGC_GET_FLAG_VALUE(EnablePrivMemNewSOATranspose);
@@ -571,9 +571,12 @@ bool IGC::SOALayoutChecker::MismatchDetected(Instruction &I) {
     return false;
 
   Type *allocaTy = allocaRef.getAllocatedType();
-  bool allocaIsVecOrArr = allocaTy->isVectorTy() || allocaTy->isArrayTy();
+  bool allocaIsVecOrArrOrStruct = allocaTy->isVectorTy() || allocaTy->isArrayTy() || allocaTy->isStructTy();
 
-  if (!allocaIsVecOrArr)
+  if (!allocaIsVecOrArrOrStruct)
+    return false;
+
+  if(mismatchedWidthsSupport)
     return false;
 
   bool useOldAlgorithm = !useNewAlgo(pInfo->baseType);
@@ -593,12 +596,24 @@ bool IGC::SOALayoutChecker::MismatchDetected(Instruction &I) {
         allocaTy = arrTy->getElementType();
       } else if (auto *vec = dyn_cast<IGCLLVM::FixedVectorType>(allocaTy)) {
         allocaTy = vec->getElementType();
+      } else if (auto *strct = dyn_cast<StructType>(allocaTy)){
+        if (auto *arrTy = dyn_cast<ArrayType>(strct->getStructElementType(0))) {
+          allocaTy = arrTy->getElementType();
+        } else if (auto *vec = dyn_cast<IGCLLVM::FixedVectorType>(strct->getStructElementType(0))){
+          allocaTy = vec->getElementType();
+        }
       }
 
       if (auto *arrTy = dyn_cast<ArrayType>(pUserTy)) {
         pUserTy = arrTy->getElementType();
       } else if (auto *vec = dyn_cast<IGCLLVM::FixedVectorType>(pUserTy)) {
         pUserTy = vec->getElementType();
+      } else if (auto *strct = dyn_cast<StructType>(pUserTy)){
+        if (auto *arrTy = dyn_cast<ArrayType>(strct->getStructElementType(0))) {
+          pUserTy = arrTy->getElementType();
+        } else if (auto *vec = dyn_cast<IGCLLVM::FixedVectorType>(strct->getStructElementType(0))){
+          pUserTy = vec->getElementType();
+        }
       }
     }
 

IGC/Compiler/CISACodeGen/LowerGEPForPrivMem.hpp

@@ -70,7 +70,7 @@ class SOALayoutChecker : public llvm::InstVisitor<SOALayoutChecker, bool> {
   friend llvm::InstVisitor<SOALayoutChecker, bool>;
 
   // isOCL is for testing, it will be removed once testing is done.
-  SOALayoutChecker(llvm::AllocaInst &allocaToCheck, bool isOCL);
+  SOALayoutChecker(llvm::AllocaInst &allocaToCheck, bool isOCL, bool mismatchedWidthsSupport=false);
   SOALayoutChecker() = delete;
   ~SOALayoutChecker() = default;
   SOALayoutChecker(SOALayoutChecker &) = delete;
@@ -87,6 +87,10 @@ class SOALayoutChecker : public llvm::InstVisitor<SOALayoutChecker, bool> {
   const llvm::DataLayout *pDL;
   std::unique_ptr<SOALayoutInfo> pInfo;
 
+  // If mismatched widths reach PrivateMemoryResolution, it should turn off optimization
+  // But if they reach LowerGEPForPrivMemPass they get resolved properly
+  bool mismatchedWidthsSupport;
+
   // ===== fields for new algo =====
   // todo: combine the new and old together
   //

IGC/Compiler/tests/PrivateMemoryResolution/SOA_promotion/soa-nested-load.ll

@@ -18,6 +18,8 @@
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024-n8:16:32"
 target triple = "spir64-unknown-unknown"
 
+%g = type { [512 x i32] }
+
 ; Function Attrs: nofree nosync nounwind
 define spir_kernel void @test(ptr nocapture writeonly %d, <8 x i32> %r0, <8 x i32> %payloadHeader, <3 x i32> %enqueuedLocalSize, i16 %localIdX, i16 %localIdY, i16 %localIdZ, ptr nocapture readnone %privateBase) {
 entry:
@@ -43,6 +45,18 @@ exit:
 ; CHECK: %load2 = load i8, ptr %arr3
   %arr3 = alloca [512 x i32]
   %load2 = load i8, ptr %arr3
+
+; This case is valid because float and i32 have 32 bits
+; CHECK: insertelement <2 x float> {{.*}}, float {{.*}}, i32 0
+; CHECK: insertelement <2 x float> {{.*}}, float {{.*}}, i32 1
+  %st = alloca %g
+  %load3 = load <2 x float>, ptr %st
+
+; This case is not valid because i32 and i8 have different sizes
+; CHECK: %load4 = load <2 x i8>, ptr %st2
+  %st2 = alloca %g
+  %load4 = load <2 x i8>, ptr %st2
+
   ret void
 }