Update on "Fix optimized log_softmax along non-contiguous dim"

#8382 certainly didn't fix this problem (and added it on x86), but I don't think it was correct on ARM prior to that either.

Added a regression test.

Differential Revision: [D69928884](https://our.internmc.facebook.com/intern/diff/D69928884/)

[ghstack-poisoned]

Author: Github Executorch

extension/flat_tensor/serialize/targets.bzl

@@ -39,7 +39,9 @@ def define_common_targets():
         name = "flat_tensor_header",
         srcs = ["flat_tensor_header.cpp"],
         exported_headers = ["flat_tensor_header.h"],
-        visibility = ["//executorch/..."],
+        visibility = [
+            "//executorch/...",
+        ],
         exported_deps = ["//executorch/runtime/core:core"],
     )
 
@@ -54,6 +56,7 @@ def define_common_targets():
         exported_headers = ["serialize.h"],
         visibility = [
             "//executorch/...",
+            "@EXECUTORCH_CLIENTS",
         ],
         exported_external_deps = ["flatbuffers-api"],
     )

kernels/test/op_log_softmax_test.cpp

@@ -77,9 +77,9 @@ class OpLogSoftmaxOutTest : public OperatorTest {
   void test_dtype_noncontiguous_dim() {
     TensorFactory<DTYPE> tf;
 
-    // clang-format off Dim 0 must be longer than the vector width of
-    // the machine (for float, this is 4 for ARM64 and 8 for AVX2) to
-    // exhibit problems.
+    // Dim 0 must be longer than the vector width of the machine (for
+    // float, this is 4 for ARM64 and 8 for AVX2) to exhibit problems.
+    // clang-format off
     Tensor x = tf.make(
       {9, 3},
       {
@@ -125,7 +125,6 @@ class OpLogSoftmaxOutTest : public OperatorTest {
       EXPECT_TENSOR_CLOSE(out, expected);
     }
   }
-
 };
 
 TEST_F(OpLogSoftmaxOutTest, Smoke) {