[Mlir-commits] [mlir] [MLIR][XeGPU] Update XeGPU create_tdesc, update_offset, load, store and prefetch. (PR #154653)

[Sang Ik Lee]

================
@@ -124,22 +122,37 @@ isValidGatherScatterParams(Type maskTy, VectorType valueTy,
 }
 
 static LogicalResult
-isValidGatherScatterBufferParams(Type maskTy, VectorType valueTy,
-                                 int64_t chunkSize,
+isValidGatherScatterBufferParams(Type offsetsTy, Type maskTy,
+                                 VectorType valueTy, int64_t chunkSize,
                                  function_ref<InFlightDiagnostic()> emitError) {
 
-  if (!valueTy)
-    return emitError() << "Expecting a vector type result.";
+  auto maskVecTy = dyn_cast<VectorType>(maskTy);
+  auto offsetsVecTy = dyn_cast<VectorType>(offsetsTy);
+  if (!valueTy) {
+    if (chunkSize > 1)
+      return emitError() << "Expecting chunk size == 1 for scalar result";
+    if (maskVecTy || offsetsVecTy)
----------------
silee2 wrote:

all of mask and offset are Vector or Scalar type
But value depends on chunk size. 
I'm not too familiar with using `llvm::all_of` and available predicate lambda to use with it.
PR is getting a bit big so will optimize in a follow up PR.

https://github.com/llvm/llvm-project/pull/154653