[Mlir-commits] [mlir] [MLIR][Conversion][XeGPU][XeVM] Add XeGPUToXeVM conversion pass and tests. (PR #154556)

[Adam Siemieniuk]

================
@@ -0,0 +1,48 @@
+// RUN: mlir-opt -convert-xegpu-to-xevm %s | FileCheck %s
+
+gpu.module @create_nd_tdesc {
+  // CHECK-LABEL: gpu.func @create_nd_tdesc
+  // CHECK-SAME: %[[ARG0:.*]]: memref<8x16xf32, 1>, %[[ARG1:.*]]: ui64
+  // CHECK-SAME: %[[ARG2:.*]]: index, %[[ARG3:.*]]: index, %[[ARG4:.*]]: index, %[[ARG5:.*]]: index
+  gpu.func @create_nd_tdesc(%src: memref<8x16xf32, 1>, %ptr: ui64, %shape1: index, %shape2: index,
+       %stride1: index, %stride2: index) kernel {
+        // CHECK: %[[VAR0:.*]] = index.castu %[[ARG1]] : ui64 to index
+        // CHECK: %[[VAR1:.*]] = arith.index_castui %[[VAR0]] : index to i64
+        // CHECK: %[[CST:.*]] = arith.constant dense<0> : vector<8xi32>
+        // CHECK: %[[C0_I32:.*]] = arith.constant 0 : i32
+        // CHECK: %[[C0_I32_0:.*]] = arith.constant 0 : i32
+        // CHECK: %[[VAR3:.*]] = arith.index_cast %[[ARG3]] : index to i32
+        // CHECK: %[[VAR5:.*]] = arith.index_cast %[[ARG2]] : index to i32
+        // CHECK: %[[VAR6:.*]] = vector.bitcast %[[CST]] : vector<8xi32> to vector<4xi64>
+        // CHECK: %[[VAR7:.*]] = vector.insert %[[VAR1]], %[[VAR6]] [0] : i64 into vector<4xi64>
+        // CHECK: %[[VAR8:.*]] = vector.bitcast %[[VAR7]] : vector<4xi64> to vector<8xi32>
+        // CHECK: %[[VAR9:.*]] = vector.insert %[[VAR3]], %[[VAR8]] [2] : i32 into vector<8xi32>
+        // CHECK: %[[VAR10:.*]] = vector.insert %[[VAR5]], %[[VAR9]] [3] : i32 into vector<8xi32>
+        // CHECK: %[[VAR11:.*]] = vector.insert %[[C0_I32]], %[[VAR10]] [4] : i32 into vector<8xi32>
+        // CHECK: %[[VAR12:.*]] = vector.insert %[[C0_I32_0]], %[[VAR11]] [5] : i32 into vector<8xi32>
+        %ptr_tdesc = xegpu.create_nd_tdesc %ptr, shape:[%shape1, %shape2], strides:[%stride1, %stride2]
+            : ui64 -> !xegpu.tensor_desc<8x16xf32>
----------------
adam-smnk wrote:

I guess it is the payload encoding.
Could you add more meaningful names to the variables?

https://github.com/llvm/llvm-project/pull/154556