[Mlir-commits] [mlir] [MLIR][XeGPU] Add intel_reqd_sub_group_size attribute to gpu.func to facilitate gpu.shuffle lowering (PR #182434)

[Sang Ik Lee]

================
@@ -628,6 +629,14 @@ void XeGPUSgToWiDistributeExperimentalPass::runOnOperation() {
       }
     });
   }
+  // iterate the IR and attach each function op with sub_group_size attribute to
+  // support shuffle lowering in later stages.
+  root->walk([&](gpu::GPUFuncOp funcOp) {
+    auto uArch = getUArch(xegpu::getChipStr(funcOp).value_or(""));
+    funcOp->setAttr("intel_reqd_sub_group_size",
----------------
silee2 wrote:

This would a considered discardable. And may not survive till later stage.

https://github.com/llvm/llvm-project/pull/182434