[Mlir-commits] [mlir] [mlir][tosa] Allow rank-0 vector operands in tosa.apply_scale (PR #199924)

[llvmlistbot at llvm.org]

================
@@ -1777,3 +1777,10 @@ func.func @test_assert_equal_shape() {
   tosa.assert_equal_shape %0, %1 {allow_broadcast = true} : (!tosa.shape<2>, !tosa.shape<2>) -> ()
   return
 }
+
+// -----
+// CHECK-LABEL: test_apply_scale_rank_0_vector
+func.func @test_apply_scale_rank_0_vector(%arg0: vector<i32>, %arg1: vector<i32>, %arg2: vector<i8>) -> vector<i32> {
----------------
ziereis wrote:

thanks for the review, i removed the tests. I tried adding a test to tosa-to-linalg however i can't get the regular linalg conversion to emit 0-d vectors, it will always just convert into scalars. Really the only way i found to get to the 0-d vector shape is with the reproducer i attached in the PR description. I think it might be a good idea to put this somewhere as test but im not sure where would be a good fit. 

https://github.com/llvm/llvm-project/pull/199924