[Mlir-commits] [mlir] [mlir][xegpu] add support for structure control flow ops in workgroup to subgroup distribution (PR #142618)

[Chao Chen]

================
@@ -169,4 +169,125 @@ gpu.func @test_dpas_no_sg_data(%a: memref<24x32xf32>, %b: memref<32x24xf32>) {
       : vector<24x32xf32>, vector<32x24xf32> -> vector<24x24xf32>
     gpu.return
   }
+
+  gpu.func @test_scf_for(%arg0: memref<1024x1024xf16>, %arg1: memref<1024x1024xf16>, %arg2: memref<1024x1024xf32>) {
+    //CHECK: [[c0:%.+]] = arith.constant 0 : index
+    //CHECK: [[c128:%.+]] = arith.constant 128 : index
+    //CHECK: [[c1024:%.+]] = arith.constant 1024 : index
+    %c0 = arith.constant 0 : index
+    %c128 = arith.constant 128 : index
+    %c1024 = arith.constant 1024 : index
+    %block_id_x = gpu.block_id  x
+    %block_id_y = gpu.block_id  y
+    %0 = arith.muli %block_id_x, %c128 : index
+    %1 = arith.muli %block_id_y, %c128 : index
+    %2 = xegpu.create_nd_tdesc %arg2[%0, %1] : memref<1024x1024xf32> -> !xegpu.tensor_desc<128x128xf32, #xegpu.layout<sg_layout = [8, 8], sg_data = [16, 16]>>
+    %3 = xegpu.load_nd %2 : !xegpu.tensor_desc<128x128xf32, #xegpu.layout<sg_layout = [8, 8], sg_data = [16, 16]>> -> vector<128x128xf32>
+    %4 = xegpu.create_nd_tdesc %arg0[%0, %c0] : memref<1024x1024xf16> -> !xegpu.tensor_desc<128x128xf16, #xegpu.layout<sg_layout = [8, 8], sg_data = [16, 128]>>
+    %5 = xegpu.create_nd_tdesc %arg1[%c0, %1] : memref<1024x1024xf16> -> !xegpu.tensor_desc<128x128xf16, #xegpu.layout<sg_layout = [8, 8], sg_data = [128, 16]>>
+
+    //CHECK: [[scf:%.+]]:3 = scf.for [[arg3:%.+]] = [[c0]] to [[c1024]] step [[c128]] iter_args([[arg4:%.+]] = {{.*}}, [[arg5:%.+]] = {{.*}}, [[arg6:%.+]] = {{.*}}) -> (!xegpu.tensor_desc<16x128xf16>, !xegpu.tensor_desc<128x16xf16>, vector<16x16xf32>)
----------------
chencha3 wrote:

Fixed

https://github.com/llvm/llvm-project/pull/142618