[Mlir-commits] [mlir] [MLIR][XeGPU] Add uArch limitation to scatter load store (PR #172845)

Charitha Saumya llvmlistbot at llvm.org
Thu Jan 8 11:03:20 PST 2026


================
@@ -215,6 +215,26 @@ struct SubgroupMatrixMultiplyAcc : public Instruction,
   const unsigned packedFormatBitSizeB;
 };
 
+struct StoreScatterInstruction : public Instruction {
+  StoreScatterInstruction()
+      : Instruction(InstructionKind::StoreScatter, InstructionScope::Lane) {}
+  static bool classof(const Instruction *B) {
+    return B->getInstructionKind() == InstructionKind::StoreScatter;
+  }
+
+  int32_t getMaxBitSize() const { return 128; }
----------------
charithaintc wrote:

what does this size mean? is it the max size that can be loaded per lane? 
nit: worth adding a small comment. 

https://github.com/llvm/llvm-project/pull/172845


More information about the Mlir-commits mailing list