Update base for Update on "[XNNPACK][Weights Cache] Initial Weights Cache Design with NamedDataMap"

XNNWeightsCache Design with NamedDataMap. The intent of the weights cache is for tensors to be loaded (via name) through the named data map.

APIs to be used by XNNCompiler:

- load_unpacked_data
    - Takes in a string name (tensor name). The weights cache loads the data for this string from the named data map and returns the pointer. It also creates a mapping of this pointer to the name which is later used by the XNNPACK's internal weight cache implementation

- free_unpacked_data
    - Frees all the unpacked data loaded from NamedDataMap. This is only safe to call after xnn_create_runtime has been called. This is because create_runtime takes unpacked data pointers and packs them into a separate buffer.

- a couple getter methods
    - get_packed_data_names
    - get_unpacked_data_names
    - get_num_packed_data
    - get() (get's the xnn_weights_cache object)


Internal APIs used by XNNPACK Library

- look_up
    - takes a cache key (weight and bias pointers) and looks up the offset to the packed weight if it exists
- look_up_or_insert
    - takes a cache key and pointer to packed weights and looks_up the offset if it exists, or inserts a new packed weight into the cache and returns that offset
- offset_to_addr
    - gets offset and returns address to packed pointer
- reserve_space
    - returns memory address with appropriate sie for XNNPACK to populate with packed weights ( I want to use the runtime_allocator for this but i don't think we have the right sizes, so for now we are just using a string buffer and resizing it)
- is_finalized
     - since this cache doesn't necessarily need to care about a finalized state we always return true.
- delete_cache
    - deletes cache

Differential Revision: [D70885917](https://our.internmc.facebook.com/intern/diff/D70885917/)

[ghstack-poisoned]

Author: mcr229

.ci/scripts/unittest-buck2.sh

@@ -17,8 +17,10 @@ buck2 query "//backends/apple/... + //backends/example/... + \
 //kernels/optimized/... + //kernels/portable/... + //kernels/quantized/... + \
 //kernels/test/... + //runtime/... + //schema/... + //test/... + //util/..."
 
+UNBUILDABLE_OPTIMIZED_OPS_REGEX="gelu|fft_r2c|log_softmax"
+BUILDABLE_OPTIMIZED_OPS=$(buck2 query //kernels/optimized/cpu/... | grep -E -v $UNBUILDABLE_OPTIMIZED_OPS_REGEX)
 # TODO: expand the covered scope of Buck targets.
 # //runtime/kernel/... is failing because //third-party:torchgen_files's shell script can't find python on PATH.
 # //runtime/test/... requires Python torch, which we don't have in our OSS buck setup.
-buck2 build //runtime/backend/... //runtime/core/... //runtime/executor: //runtime/kernel/... //runtime/platform/...
-buck2 test //runtime/backend/... //runtime/core/... //runtime/executor: //runtime/kernel/... //runtime/platform/...
+buck2 test $BUILDABLE_OPTIMIZED_OPS //kernels/portable/... //runtime/backend/... //runtime/core/... \
+      //runtime/executor: //runtime/kernel/... //runtime/platform/...

.lintrunner.toml

@@ -218,6 +218,8 @@ exclude_patterns = [
     'examples/**',
     'extension/**',
     'kernels/optimized/**',
+    # Justified <functional> include.
+    'runtime/kernel/thread_parallel_interface.h',
     'scripts/**',
     'third-party/**',
     'util/**',

CMakeLists.txt

@@ -751,7 +751,6 @@ if(EXECUTORCH_BUILD_PTHREADPOOL
    AND EXECUTORCH_BUILD_CPUINFO
 )
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/threadpool)
-  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/parallel)
 endif()
 
 if(EXECUTORCH_BUILD_PYBIND)

CODEOWNERS

@@ -52,31 +52,31 @@
 /extension/export_util @kimishpatel
 /extension/flat_tensor @lucylq
 /extension/gguf_util @larryliu0820
-/extension/kernel_util @kimishpatel @manuelcandales
-/extension/llm @jackzhxng @iseeyuan @larryliu0820
-/extension/memory_allocator @JacobSzwejbka
+/extension/kernel_util @kimishpatel @manuelcandales @swolchok
+/extension/llm @jackzhxng @iseeyuan @larryliu0820 @swolchok
+/extension/memory_allocator @JacobSzwejbka @swolchok
 /extension/module @shoumikhin
-/extension/parallel @kimishpatel
+/extension/parallel @kimishpatel @swolchok
 /extension/pybindings @JacobSzwejbka @larryliu0820
-/extension/pytree @JacobSzwejbka
-# /extension/runner_util @dbort
+/extension/pytree @JacobSzwejbka @swolchok
+/extension/runner_util @swolchok
 /extension/tensor @shoumikhin
-# /extension/testing_util @dbort
-/extension/threadpool @kimishpatel
+/extension/testing_util @swolchok
+/extension/threadpool @kimishpatel @swolchok
 /extension/training @JacobSzwejbka
 
-/kernels @manuelcandales
+/kernels @manuelcandales @swolchok
 
 /profiler @tarun292 @Gasoonjia
 
-/runtime @JacobSzwejbka @lucylq
+/runtime @JacobSzwejbka @lucylq @swolchok
 /runtime/backend @cccclai
 
 /schema @JacobSzwejbka @lucylq
 
-/scripts @GregoryComer
+/scripts @GregoryComer @swolchok
 
-/shim @larryliu0820 @GregoryComer
+/shim @larryliu0820 @GregoryComer @swolchok
 
 /third-party @GregoryComer
 

Test.cmake

@@ -13,7 +13,6 @@ if(BUILD_TESTING)
   add_subdirectory(extension/evalue_util/test)
   add_subdirectory(extension/kernel_util/test)
   add_subdirectory(extension/memory_allocator/test)
-  add_subdirectory(extension/parallel/test)
   add_subdirectory(extension/pytree/test)
   add_subdirectory(kernels/portable/cpu/util/test)
   add_subdirectory(kernels/prim_ops/test)

backends/arm/test/conftest.py

@@ -32,7 +32,10 @@ def pytest_configure(config):
     pytest._test_options = {}  # type: ignore[attr-defined]
     pytest._test_options["corstone_fvp"] = False  # type: ignore[attr-defined]
 
-    if config.option.arm_run_corstoneFVP:
+    if (
+        getattr(config.option, "arm_run_corstoneFVP", False)
+        and config.option.arm_run_corstoneFVP
+    ):
         corstone300_exists = shutil.which("FVP_Corstone_SSE-300_Ethos-U55")
         corstone320_exists = shutil.which("FVP_Corstone_SSE-320")
         if not (corstone300_exists and corstone320_exists):

backends/cadence/aot/TARGETS

@@ -115,11 +115,23 @@ python_library(
     ],
     deps = [
         "fbcode//caffe2:torch",
-        "fbcode//executorch/exir:scalar_type",
         "fbcode//executorch/backends/cadence/aot:utils",
     ],
 )
 
+python_library(
+    name = "ref_implementations",
+    srcs = [
+        "ref_implementations.py",
+    ],
+    typing = True,
+    deps = [
+        "fbcode//caffe2:torch",
+        "fbcode//executorch/exir:scalar_type",
+    ],
+)
+
+
 export_file(name = "functions.yaml")
 
 executorch_generated_lib(

backends/cadence/aot/compiler.py

@@ -198,6 +198,8 @@ def export_to_edge(
             _skip_dim_order=True,
             # Allow specific non-core aten ops in the IR.
             _core_aten_ops_exception_list=[
+                torch.ops.aten._linalg_det.default,
+                torch.ops.aten._linalg_svd.default,
                 torch.ops.aten._native_batch_norm_legit_functional.default,
                 torch.ops.aten.linear.default,
                 torch.ops.aten.linalg_vector_norm.default,

backends/cadence/aot/export_example.py

@@ -38,6 +38,8 @@ def export_model(
     example_inputs: Tuple[Any, ...],
     file_name: str = "CadenceDemoModel",
     run_and_compare: bool = True,
+    eps_error: float = 1e-1,
+    eps_warn: float = 1e-5,
 ):
     # create work directory for outputs and model binary
     working_dir = tempfile.mkdtemp(dir="/tmp")
@@ -89,4 +91,6 @@ def export_model(
             inputs=example_inputs,
             ref_outputs=ref_outputs,
             working_dir=working_dir,
+            eps_error=eps_error,
+            eps_warn=eps_warn,
         )

backends/cadence/aot/functions.yaml

@@ -248,3 +248,8 @@
   kernels:
     - arg_meta: null
       kernel_name: impl::reference::quantized_fully_connected_per_tensor_out
+
+- func: cadence::requantize.out(Tensor input, Tensor in_scale, Tensor in_zero_point, Tensor out_scale, Tensor out_zero_point, ScalarType out_dtype, *, Tensor(a!) out) -> Tensor(a!)
+  kernels:
+    - arg_meta: null
+      kernel_name: impl::reference::requantize_out