[Mlir-commits] [mlir] [SCF] Fixed epilogue predicates in loop pipelining (PR #108964)

llvmlistbot at llvm.org llvmlistbot at llvm.org
Wed Sep 18 11:30:32 PDT 2024


================
@@ -802,8 +801,10 @@ func.func @stage_0_value_escape(%A: memref<?xf32>, %result: memref<?xf32>, %ub:
 //        CHECK:   } else {
 //        CHECK:   }
 //        CHECK:   return
-func.func @dynamic_loop(%A: memref<?xf32>, %result: memref<?xf32>, %lb: index, %ub: index, %step: index) {
+func.func @dynamic_loop(%A: memref<?xf32>, %result: memref<?xf32>, %ub: index) {
   %cf = arith.constant 1.0 : f32
+  %lb = arith.constant 3 : index
+  %step = arith.constant 2 : index
----------------
sjw36 wrote:

It helps demonstrates that the `iterI >= 0` is not related to the lower_bound.

https://github.com/llvm/llvm-project/pull/108964


More information about the Mlir-commits mailing list