[Mlir-commits] [mlir] [mlir][linalg] Guard pack tensor semantics (PR #206011)

[Adam Siemieniuk]

================
@@ -0,0 +1,17 @@
+// RUN: mlir-opt -sparsification-and-bufferization -transform-interpreter -split-input-file -verify-diagnostics -allow-unregistered-dialect %s
+
+module attributes {gpu.data_layout = "chunked", impl.libfunc.name = "test.matmul"} {
+  func.func @matmul(%arg0: tensor<?x?xf32>, %arg1: tensor<?x?xf32>, %arg2: tensor<?x?xf32>) -> tensor<?x?xf32> {
+    %0 = linalg.matmul ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?x?xf32>) outs(%arg2 : tensor<?x?xf32>) -> tensor<?x?xf32>
----------------
adam-smnk wrote:

Please rewrite it into matmul on memrefs to narrow the test's scope to the pack transform only.

https://github.com/llvm/llvm-project/pull/206011