Qualcomm AI Engine Direct - Fix Missing Kernels (pytorch#2040)

Summary:
- Complie and link to aten kernels
- Add LLama partition number test cases

Pull Request resolved: pytorch#2040

Reviewed By: mergennachin

Differential Revision: D54071603

Pulled By: cccclai

fbshipit-source-id: 3731bc4a9e14c555e79adfddc6e648669d3da7d0

backends/qualcomm/tests/test_qnn_delegate.py

@@ -24,6 +24,7 @@
 from executorch.examples.models.edsr import EdsrModel
 from executorch.examples.models.inception_v3 import InceptionV3Model
 from executorch.examples.models.inception_v4 import InceptionV4Model
+from executorch.examples.models.llama2 import Llama2Model
 from executorch.examples.models.mobilebert import MobileBertModelExample
 from executorch.examples.models.mobilenet_v2 import MV2Model
 from executorch.examples.models.mobilenet_v3 import MV3Model
@@ -421,6 +422,7 @@ def test_qnn_backend_example_models(self):
             EdsrModel(),
             InceptionV3Model(),
             InceptionV4Model(),
+            Llama2Model(),
             MV2Model(),
             MV3Model(),
             MobileBertModelExample(),
@@ -437,6 +439,7 @@ def test_qnn_backend_example_models(self):
             1,
             1,
             1,
+            1,
         ]
         # TODO: Due to trigger maximum recursion depth exceeded, need to check it.
         disable_validation()
@@ -895,6 +898,7 @@ def test_qnn_backend_example_models(self):
             {"module": EdsrModel(), "annotation": (annotate_forward,)},
             {"module": InceptionV3Model(), "annotation": ()},
             {"module": InceptionV4Model(), "annotation": ()},
+            {"module": Llama2Model(), "annotation": ()},
             {"module": MV2Model(), "annotation": ()},
             {"module": MV3Model(), "annotation": ()},
             # only works on QNN 2.12 so far
@@ -909,6 +913,7 @@ def test_qnn_backend_example_models(self):
             1,
             1,
             1,
+            1,
             # For MobileBertModelExample
             # 1,
             1,

examples/qualcomm/CMakeLists.txt

@@ -55,8 +55,8 @@ gen_selected_ops("" "" "ON")
 generate_bindings_for_kernels(
   ${EXECUTORCH_ROOT}/kernels/portable/functions.yaml ""
 )
-gen_operators_lib("portable_ops_lib" portable_kernels executorch)
-target_include_directories(portable_ops_lib
+gen_operators_lib("full_portable_ops_lib" portable_kernels executorch)
+target_include_directories(full_portable_ops_lib
     PUBLIC
     ${_common_include_directories}
 )
@@ -87,7 +87,7 @@ target_include_directories(qnn_executor_runner
 )
 target_link_libraries(qnn_executor_runner
     qnn_executorch_backend
-    portable_ops_lib
+    full_portable_ops_lib
     gflags
 )
 target_compile_options(qnn_executor_runner