[Mlir-commits] [mlir] [MLIR][XeGPU] Add 2D `vector.multi_reduction` optimization (PR #171154)

[Charitha Saumya]

================
@@ -77,6 +77,7 @@ void buildGPUPassPipeline(OpPassManager &pm,
     layoutOptions.layoutKind = "lane";
     pm.addNestedPass<gpu::GPUModuleOp>(
         xegpu::createXeGPUPropagateLayout(layoutOptions));
+    pm.addNestedPass<gpu::GPUModuleOp>(xegpu::createXeGPUOptimizeBlockLoads());
----------------
charithaintc wrote:

yeah we need to call it again. after optimize block loads we need to run canonicalize and cse which can add some new vector instructions. 

We can avoid running canonicalize and cse, but it may need some more code changes (I would prefer to avoid changing code for this). so better to rerun.  

https://github.com/llvm/llvm-project/pull/171154