[Mlir-commits] [mlir] [mlir][xegpu] Add XeGPU subgroup map propagation analysis for XeGPU SIMT distribution. (PR #130240)

[Jianhui Li]

================
@@ -0,0 +1,563 @@
+// RUN: mlir-opt -xegpu-subgroup-distribute='print-analysis-only=true' -split-input-file %s | FileCheck %s
+
+// CHECK: function: test_dpas_f16:
+// CHECK-NEXT: argument: <block argument> of type 'memref<8x16xf16>' at index: 0
+// CHECK-NEXT: sg_map  : Not assigned.
+// CHECK-NEXT: argument: <block argument> of type 'memref<16x16xf16>' at index: 1
+// CHECK-NEXT: sg_map  : Not assigned.
+// CHECK-NEXT: argument: <block argument> of type 'memref<8x16xf32>' at index: 2
+// CHECK-NEXT: sg_map  : Not assigned.
+// CHECK-NEXT: op    : %{{.*}} = arith.constant 0 : index
+// CHECK-NEXT: sg_map for result #0: Not assigned.
+// CHECK-NEXT: op    : %{{.*}} = arith.constant dense<0.000000e+00> : vector<8x16xf32>
+// CHECK-NEXT: sg_map for result #0: wi_layout: [1, 16], wi_data: [1, 1]
+// CHECK-NEXT: op    : %[[T0:.*]] = xegpu.create_nd_tdesc %{{.*}} : memref<8x16xf16> -> !xegpu.tensor_desc<8x16xf16>
+// CHECK-NEXT: sg_map for result #0: wi_layout: [1, 16], wi_data: [1, 1]
----------------
Jianhui-Li wrote:

what is this map for? 

https://github.com/llvm/llvm-project/pull/130240