Disable vectorization for atomic_cas on xpu backend

Signed-off-by: Witold Dziurdz <[email protected]>

Author: wdziurdz

third_party/intel/lib/TritonIntelGPUToLLVM/LoadStoreOpToLLVM.cpp

@@ -3154,36 +3154,21 @@ struct AtomicCASOpConversion
                  : valueTy;
     auto valueElemNBits = valueElemTy.getIntOrFloatBitWidth();
     auto elemsPerThread = getTotalElemsPerThread(op.getVal().getType());
-    // vec = 1 for scalar
-    auto vec = getVectorSize(op.getPtr());
-    // tensor
-    if (tensorTy) {
-      auto valTy = cast<RankedTensorType>(op.getVal().getType());
-      vec = std::min<unsigned>(vec, valTy.getElementType().isF16() ? 2 : 1);
-    }
 
     auto freeVarMasks = getFreeVariableMasks(valueTy);
     Value mask =
         emitRedundantThreadPredicate(freeVarMasks, rewriter, loc, targetInfo);
-    auto vecTy = vec_ty(valueElemTy, vec);
     SmallVector<Value> resultVals(elemsPerThread);
 
     MemSemantic memSem = op.getSem();
     LLVM::AtomicOrdering successOrdering = getMemoryOrdering(memSem)
                                                ? *getMemoryOrdering(memSem)
                                                : LLVM::AtomicOrdering::acq_rel;
     LLVM::AtomicOrdering failureOrdering = LLVM::AtomicOrdering::monotonic;
-    for (size_t i = 0; i < elemsPerThread; i += vec) {
-      Value casVal = b.undef(vecTy);
-      for (int ii = 0; ii < vec; ++ii) {
-        Value iiVal = createIndexAttrConstant(
-            rewriter, loc, getTypeConverter()->getIndexType(), ii);
-        casVal = b.insert_element(vecTy, casVal, valElements[i + ii], iiVal);
-      }
-
+    for (size_t i = 0; i < elemsPerThread; ++i) {
       Value casPtr = ptrElements[i];
       Value casCmp = cmpElements[i];
-      casVal = valElements[i];
+      Value casVal = valElements[i];
 
       assert((valueElemNBits == 32 || valueElemNBits == 64) &&
              "Unexpected width");
@@ -3212,15 +3197,11 @@ struct AtomicCASOpConversion
       } else {
         ret = createAtomicCASInstruction()[0];
       }
-      Type retType = (!tensorTy || vec == 1) ? valueElemTy : vecTy;
-      ret = b.bitcast(ret, retType);
+
+      ret = b.bitcast(ret, valueElemTy);
 
       if (tensorTy) {
-        for (int ii = 0; ii < vec; ++ii) {
-          resultVals[i + ii] =
-              vec == 1 ? ret
-                       : b.extract_element(valueElemTy, ret, b.i32_val(ii));
-        }
+        resultVals[i] = ret;
       } else {
         if (op.getResult().use_empty()) {
           rewriter.eraseOp(op);