[LLVM16][PrivateMemoryResolution] Adding another case of mismatch detection

Current implementation of MismatchDetection was skipping cases due to algorithm version check.
It prevented mismatched cases to be detected, so I'm removing that check.

Author: bokrzesi

IGC/Compiler/CISACodeGen/LowerGEPForPrivMem.cpp

@@ -576,40 +576,37 @@ bool IGC::SOALayoutChecker::MismatchDetected(Instruction &I) {
   if (!allocaIsVecOrArr)
     return false;
 
-  bool useOldAlgorithm = !useNewAlgo(pInfo->baseType);
+  auto DL = I.getParent()->getParent()->getParent()->getDataLayout();
 
-  if (useOldAlgorithm) {
-    auto DL = I.getParent()->getParent()->getParent()->getDataLayout();
+  Type *pUserTy = I.getType();
 
-    Type *pUserTy = I.getType();
+  if (auto *storeInst = dyn_cast<StoreInst>(&I))
+    pUserTy = storeInst->getValueOperand()->getType();
 
-    if (auto *storeInst = dyn_cast<StoreInst>(&I))
-      pUserTy = storeInst->getValueOperand()->getType();
-
-    if (auto *pgep = dyn_cast<GetElementPtrInst>(parentLevelInst)) {
-      allocaTy = pgep->getResultElementType();
-    } else {
-      if (auto *arrTy = dyn_cast<ArrayType>(allocaTy)) {
-        allocaTy = arrTy->getElementType();
-      } else if (auto *vec = dyn_cast<IGCLLVM::FixedVectorType>(allocaTy)) {
-        allocaTy = vec->getElementType();
-      }
+  if (auto *pgep = dyn_cast<GetElementPtrInst>(parentLevelInst)) {
+    allocaTy = pgep->getResultElementType();
+  } else {
+    if (auto *arrTy = dyn_cast<ArrayType>(allocaTy)) {
+      allocaTy = arrTy->getElementType();
+    } else if (auto *vec = dyn_cast<IGCLLVM::FixedVectorType>(allocaTy)) {
+      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();
-      }
+    if (auto *arrTy = dyn_cast<ArrayType>(pUserTy)) {
+      pUserTy = arrTy->getElementType();
+    } else if (auto *vec = dyn_cast<IGCLLVM::FixedVectorType>(pUserTy)) {
+      pUserTy = vec->getElementType();
     }
+  }
 
-    auto allocaSize = DL.getTypeAllocSize(allocaTy);
-    auto vecTySize = DL.getTypeAllocSize(pUserTy);
+  auto allocaSize = DL.getTypeAllocSize(allocaTy);
+  auto vecTySize = DL.getTypeAllocSize(pUserTy);
 
-    if (vecTySize != allocaSize) {
-      pInfo->canUseSOALayout = false;
-      return true;
-    }
+  if (vecTySize != allocaSize) {
+    pInfo->canUseSOALayout = false;
+    return true;
   }
+
   return false;
 }
 

IGC/Compiler/tests/PrivateMemoryResolution/SOA_promotion/soa-nested-load.ll renamed to IGC/Compiler/tests/PrivateMemoryResolution/SOA_promotion/soa-mismatch-detection.ll

@@ -10,14 +10,15 @@
 ;
 ; RUN: igc_opt --ocl --igc-private-mem-resolution --regkey "EnablePrivMemNewSOATranspose=1,EnableOpaquePointersBackend=1" -S %s | FileCheck %s
 ;
-; In JointMatrix SYCL test
-; https://github.com/intel/llvm/blob/sycl/sycl/test-e2e/Matrix/joint_matrix_bf16_fill_k_cache_unroll.cpp
-; I've found that such assert was failing info: error, assertion failed: bits == elementSize
-; The purpose of this test is to validate whether alloca->gep->load is not crashing
+; This test is testing "MismatchDetected" algorithm in LowerGEPForPrivMem.cpp
+; The purpose of this test is to validate whether various combinations of allocas/geps/load/stores
+; are not mismatched and then failing due to asserts or causing miscalculations at runtime.
 
 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"
 
+%"struct.ispc::vec_t" = type { i32, i32, 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 +44,18 @@ exit:
 ; CHECK: %load2 = load i8, ptr %arr3
   %arr3 = alloca [512 x i32]
   %load2 = load i8, ptr %arr3
+
+; Case Alloca->Store->Gep->Store: This case is not valid due to different sizes
+
+; CHECK:    store <4 x i32> zeroinitializer, ptr %offset.i.i.i.i.privateBufferPTR
+; CHECK:    %offset_gep = getelementptr i8, ptr %offset.i.i.i.i.privateBufferPTR, i32 16
+; CHECK:    store i32 0, ptr %offset_gep, align 4
+
+  %offset.i.i.i.i = alloca [8 x %"struct.ispc::vec_t"], align 4
+  store <4 x i32> zeroinitializer, ptr %offset.i.i.i.i, align 4
+  %offset_gep = getelementptr i8, ptr %offset.i.i.i.i, i32 16
+  store i32 0, ptr %offset_gep, align 4
+
   ret void
 }