Update on "[ET-VK] Introduce add_tensor overloads consuming TensorRef"

From ssjia:
> we should always make sure to store references produced from `graph.get_val()` only after any calls to `graph.add_*()` (i.e. modifications to the values list) are made. This is because `graph.values_`, being a `std::vector`, will reallocate with more space and move its contents if the current allocation is not sufficient. This means that if you store a reference then call `graph.add_*()` then the underlying resource the reference points to may have been moved.

I think we can guard against this behavior by passing a `TensorRef` directly, and never having to declare a variable `TensorRef& tref` in the caller's scope. An example is shown in `Staging.cpp`. We could have it consume `ValueRef` for brevity of the passing parameter but IMO it hinders readability.

Differential Revision: [D55703483](https://our.internmc.facebook.com/intern/diff/D55703483/)

[ghstack-poisoned]

Author: Jorge Pineda

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-a52b4e22571507abc35c2d47de138497190d2e0a
+0a038cf0cff2d071b7359ac0491fd2ba7798a438

.ci/scripts/test_llama.sh

@@ -118,7 +118,7 @@ EXPORTED_MODEL_NAME="${EXPORTED_MODEL_NAME}.pte"
 echo "Exporting ${EXPORTED_MODEL_NAME}"
 EXPORT_ARGS="-c stories110M.pt -p ${PARAMS} -d ${DTYPE} -n ${EXPORTED_MODEL_NAME}"
 if [[ "${MODE}" == "xnnpack" ]]; then
-  EXPORT_ARGS="${EXPORT_ARGS} --pt2e_quantize xnnpack_dynamic"
+  EXPORT_ARGS="${EXPORT_ARGS} -kv --use_sdpa_with_kv_cache -X -qmode 8da4w -G 128"
 fi
 $PYTHON_EXECUTABLE -m examples.models.llama2.export_llama ${EXPORT_ARGS}
 

.github/workflows/_unittest.yml

@@ -37,6 +37,7 @@ jobs:
         # Setup MacOS dependencies as there is no Docker support on MacOS atm
         PYTHON_EXECUTABLE=python \
         EXECUTORCH_BUILD_PYBIND=ON \
+        EXECUTORCH_BUILD_XNNPACK=ON \
         .ci/scripts/setup-linux.sh "${BUILD_TOOL}"
 
         # Run pytest with coverage

.github/workflows/apple.yml

@@ -100,6 +100,17 @@ jobs:
           zip -r "${RUNNER_TEMP}/artifacts/${FRAMEWORK}-${VERSION}.zip" "${FRAMEWORK}.xcframework"
         ) done
 
+        # Build Debug iOS Frameworks
+        PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \
+        build/build_apple_frameworks.sh --coreml --custom --mps --optimized --portable --quantized --xnnpack --Debug
+
+        # Bundle Debug iOS Frameworks
+        for FRAMEWORK in "${FRAMEWORKS[@]}"; do (
+          cd cmake-out && \
+          mv "${FRAMEWORK}.xcframework" "${FRAMEWORK}_debug.xcframework" && \
+          zip -r "${RUNNER_TEMP}/artifacts/${FRAMEWORK}_debug-${VERSION}.zip" "${FRAMEWORK}_debug.xcframework"
+        ) done
+
         popd
 
   upload-frameworks-ios:
@@ -146,5 +157,6 @@ jobs:
 
           for FILENAME in "${RUNNER_TEMP}"/frameworks-ios/*.zip; do
             [ -e "${FILENAME}" ] || continue
+            shasum -a 256 "${FILENAME}"
             ${AWS_CMD} "${FILENAME}" s3://ossci-ios/executorch/ --acl public-read
           done

.github/workflows/pull.yml

@@ -232,6 +232,7 @@ jobs:
         # build module for executorch.extension.pybindings.portable_lib
         BUILD_TOOL=${{ matrix.build-tool }}
         PYTHON_EXECUTABLE=python \
+        EXECUTORCH_BUILD_XNNPACK=ON \
         EXECUTORCH_BUILD_PYBIND=ON \
         bash .ci/scripts/setup-linux.sh "${BUILD_TOOL}"
 

.gitmodules

@@ -37,9 +37,6 @@
 [submodule "third-party/flatcc"]
 	path = third-party/flatcc
 	url = https://github.com/dvidelabs/flatcc.git
-[submodule "examples/third-party/llama"]
-	path = examples/third-party/llama
-	url = https://github.com/facebookresearch/llama.git
 [submodule "examples/third-party/fbjni"]
 	path = examples/third-party/fbjni
 	url = https://github.com/facebookincubator/fbjni.git