Only error out when the number of kernels being registered exceeds th…

…e limit (pytorch#1487)

Summary:
Pull Request resolved: pytorch#1487

As titled. In some of the CI failures such as this one: https://github.com/pytorch/executorch/actions/runs/7331918023/job/19965267429?pr=1485#step:11:1520

The error message is confusing:
```
The total number of kernels to be registered is larger than the limit 17. 15 kernels are already registered and we're trying to register another 2 kernels.
```

Here I'm changing the ">=" condition to be ">" so that the error message makes sense.

Reviewed By: lucylq

Differential Revision: D52422427

fbshipit-source-id: 645d38bb1b52f27e525016984552a29f06e54ac0

runtime/kernel/operator_registry.cpp

@@ -41,7 +41,7 @@ Error OperatorRegistry::register_kernels(const ArrayRef<Kernel>& kernels) {
   // have any side effect even if falled multiple times.
   ::et_pal_init();
 
-  if (kernels.size() + this->num_kernels_ >= kMaxNumOfKernels) {
+  if (kernels.size() + this->num_kernels_ > kMaxNumOfKernels) {
     ET_LOG(
         Error,
         "The total number of kernels to be registered is larger than the limit %" PRIu32

runtime/kernel/targets.bzl

@@ -24,7 +24,7 @@ def define_common_targets():
     )
 
     runtime.cxx_library(
-        name = "operator_registry_TWO_KERNELS_TEST_ONLY",
+        name = "operator_registry_MAX_NUM_KERNELS_TEST_ONLY",
         srcs = ["operator_registry.cpp"],
         exported_headers = ["operator_registry.h"],
         visibility = [
@@ -35,7 +35,7 @@ def define_common_targets():
             "//executorch/runtime/core:core",
             "//executorch/runtime/core:evalue",
         ],
-        preprocessor_flags = ["-DMAX_KERNEL_NUM=2"],
+        preprocessor_flags = ["-DMAX_KERNEL_NUM=1"],
     )
 
     for aten_mode in (True, False):

runtime/kernel/test/operator_registry_max_kernel_num_test.cpp

@@ -27,7 +27,7 @@ class OperatorRegistryMaxKernelNumTest : public ::testing::Test {
   }
 };
 
-// Register one kernel when max_kernel_num=2; success
+// Register one kernel when max_kernel_num=1; success
 TEST_F(OperatorRegistryMaxKernelNumTest, RegisterOneOp) {
   Kernel kernels[] = {Kernel("foo", [](RuntimeContext&, EValue**) {})};
   ArrayRef<Kernel> kernels_array = ArrayRef<Kernel>(kernels);
@@ -37,13 +37,15 @@ TEST_F(OperatorRegistryMaxKernelNumTest, RegisterOneOp) {
   EXPECT_TRUE(hasOpsFn("foo"));
 }
 
-// Register two kernels when max_kernel_num=2; fail
+// Register two kernels when max_kernel_num=1; fail
 TEST_F(OperatorRegistryMaxKernelNumTest, RegisterTwoOpsFail) {
   Kernel kernels[] = {
       Kernel("foo1", [](RuntimeContext&, EValue**) {}),
       Kernel("foo2", [](RuntimeContext&, EValue**) {})};
   ArrayRef<Kernel> kernels_array = ArrayRef<Kernel>(kernels);
-  ET_EXPECT_DEATH({ register_kernels(kernels_array); }, "");
+  ET_EXPECT_DEATH(
+      { register_kernels(kernels_array); },
+      "The total number of kernels to be registered is larger than the limit 1");
 }
 
 } // namespace executor

runtime/kernel/test/targets.bzl

@@ -25,7 +25,7 @@ def define_common_targets():
             "operator_registry_max_kernel_num_test.cpp",
         ],
         deps = [
-            "//executorch/runtime/kernel:operator_registry_TWO_KERNELS_TEST_ONLY",
+            "//executorch/runtime/kernel:operator_registry_MAX_NUM_KERNELS_TEST_ONLY",
             "//executorch/runtime/kernel:kernel_runtime_context",
         ],
     )