[Mlir-commits] [mlir] [MLIR][LLVM] Add Continuous Loop Peeling transform to SCF (PR #71555)

[Thomas Raoux]

================
@@ -0,0 +1,98 @@
+// RUN: mlir-opt %s --transform-interpreter -split-input-file | FileCheck %s
+
+#map = affine_map<(d0) -> ()>
+#map1 = affine_map<(d0) -> (d0)>
+module {
+  func.func @foo(%arg0: f32, %arg1: tensor<?xf32>) -> tensor<?xf32> {
+    %0 = linalg.generic {indexing_maps = [#map, #map1], iterator_types = ["parallel"]} ins(%arg0 : f32) outs(%arg1 : tensor<?xf32>) {
+    ^bb0(%in: f32, %out: f32):
+      %3 = arith.mulf %in, %out : f32
+      linalg.yield %3 : f32
+    } -> tensor<?xf32>
+    return %0 : tensor<?xf32>
----------------
ThomasRaoux wrote:

Can you make the test more elementary? Ideally it should start directly from a loop and not depend on linalg. This will avoid dependency on tiling/linalg that are completely independent of this.

https://github.com/llvm/llvm-project/pull/71555