[Mlir-commits] [mlir] [mlir][xegpu] Remove chunk_size attribute; infer from types (PR #205122)

[Jianhui Li]

================
@@ -1892,17 +1891,16 @@ static xegpu::DistributeLayoutAttr setupGenericLoadAnchorLayout(
 
 /// Sets up the anchor layout for a load gather operation.
 xegpu::DistributeLayoutAttr xegpu::setupLoadGatherAnchorLayout(
-    xegpu::LayoutKind layoutKind, VectorType resVecTy, int contigChunkSize,
+    xegpu::LayoutKind layoutKind, VectorType resVecTy,
     xegpu::DistributeLayoutAttr consumerLayout, const uArch::uArch *uArch) {
 
   const int subgroupSize = uArch->getSubgroupSize();
   ArrayRef<int64_t> resShape = resVecTy.getShape();
   auto context = resVecTy.getContext();
 
-  const auto *uArchInstruction = dyn_cast<xegpu::uArch::LoadGatherInstruction>(
-      uArch->getInstruction(xegpu::uArch::InstructionKind::LoadGather));
-  int maxChunkSize =
-      std::min(uArchInstruction->getMaxLaneAccessSizeBytes(), contigChunkSize);
+  // Gather loads one element per lane, so the innermost per-lane vector is
+  // capped at a single element.
+  int maxChunkSize = 1;
----------------
Jianhui-Li wrote:

I dont' understand why gather load just load one element per lane? 

https://github.com/llvm/llvm-project/pull/205122