Update on "[ET-VK][ez] Retry with no optimization during shader compilation"

## Context

In some cases, a shader can be compiled successfully with `glslc`, but if an optimization flag such as `-O` or `-Os` is used, the compilation will fail with a message similar to

```
internal error: compilation succeeded but failed to optimize: Invalid use of 8- or 16-bit result\n  %133 = OpExtInst %half %1 Pow %125 %132
```

Which will cause the shader compilation script to fail.

This diff introduces a check that if `glslc` complains about not being able to optimize the SPIR-V, then retry compilation without optimization arguments.

Differential Revision: [D70906097](https://our.internmc.facebook.com/intern/diff/D70906097/)

[ghstack-poisoned]

Author: SS-JIA

.buckconfig

@@ -39,3 +39,6 @@
 
 [buck2]
 restarter=true
+
+[oss]
+folly_cxx_tests = False

.ci/scripts/build_llama_android.sh

@@ -42,6 +42,10 @@ install_executorch_and_backend_lib() {
 
 build_llama_runner() {
     echo "Building llama runner for Android..."
+    pushd extension/llm/tokenizers
+    echo "Updating tokenizers submodule"
+    git submodule update --init
+    popd
     ANDROID_ABI=arm64-v8a
     cmake -DBUCK2="${BUCK2}" \
     -DCMAKE_TOOLCHAIN_FILE="$ANDROID_NDK"/build/cmake/android.toolchain.cmake  \

.ci/scripts/gather_test_models.py

@@ -30,6 +30,7 @@
         "dl3": "linux.4xlarge.memory",
         "emformer_join": "linux.4xlarge.memory",
         "emformer_predict": "linux.4xlarge.memory",
+        "phi-4-mini": "linux.4xlarge.memory",
     }
 }
 

.ci/scripts/test_ane_static_llama.sh

@@ -17,6 +17,12 @@ fi
 
 which "${PYTHON_EXECUTABLE}"
 
+# Update tokenizers submodule
+pushd $EXECUTORCH_ROOT/extension/llm/tokenizers
+echo "Update tokenizers submodule"
+git submodule update --init
+popd
+
 pushd $EXECUTORCH_ROOT/examples/apple/coreml/llama
 
 # Download stories llama110m artifacts

.ci/scripts/test_llama.sh

@@ -173,6 +173,10 @@ cmake_install_executorch_libraries() {
 
 cmake_build_llama_runner() {
     echo "Building llama runner"
+    pushd extension/llm/tokenizers
+    echo "Updating tokenizers submodule"
+    git submodule update --init
+    popd
     dir="examples/models/llama"
     retry cmake \
         -DCMAKE_INSTALL_PREFIX=cmake-out \

.ci/scripts/unittest-buck2.sh

@@ -8,11 +8,17 @@ set -eux
 
 # TODO: expand this to //...
 # TODO: can't query cadence & vulkan backends
+# TODO: can't query //kernels/prim_ops because of a cpp_unittest and
+# broken code in shim to read oss.folly_cxx_tests. Sending fix but it
+# needs to propagate and we need a submodule update.
 buck2 query "//backends/apple/... + //backends/example/... + \
 //backends/mediatek/... + //backends/test/... + //backends/transforms/... + \
-//backends/xnnpack/... + //configurations/... + //kernels/portable/cpu/... + \
-//runtime/... + //schema/... + //test/... + //util/..."
+//backends/xnnpack/... + //configurations/... + //kernels/aten/... + \
+//kernels/optimized/... + //kernels/portable/... + //kernels/quantized/... + \
+//kernels/test/... + //runtime/... + //schema/... + //test/... + //util/..."
 
 # TODO: expand the covered scope of Buck targets.
-buck2 build //runtime/core/portable_type/...
-buck2 test //runtime/core/portable_type/...
+# //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/...

.github/workflows/doc-build.yml

@@ -83,6 +83,7 @@ jobs:
         cp -rf docs/_build/html/* "${RUNNER_DOCS_DIR}"
 
         mv docs/_build/html "${RUNNER_ARTIFACT_DIR}"
+        cp -rf "${RUNNER_DOCS_DIR}"/javadoc "${RUNNER_ARTIFACT_DIR}"/html
 
         ls -R "${RUNNER_ARTIFACT_DIR}"/*/*.html
 

.gitmodules

@@ -28,15 +28,9 @@
 [submodule "backends/xnnpack/third-party/pthreadpool"]
 	path = backends/xnnpack/third-party/pthreadpool
 	url = https://github.com/Maratyszcza/pthreadpool.git
-[submodule "extension/llm/third-party/abseil-cpp"]
-	path = extension/llm/third-party/abseil-cpp
-	url = https://github.com/abseil/abseil-cpp.git
-[submodule "extension/llm/third-party/re2"]
-	path = extension/llm/third-party/re2
-	url = https://github.com/google/re2.git
-[submodule "extension/llm/third-party/sentencepiece"]
-	path = extension/llm/third-party/sentencepiece
-	url = https://github.com/google/sentencepiece.git
+[submodule "extension/llm/tokenizers"]
+	path = extension/llm/tokenizers
+	url = https://github.com/pytorch-labs/tokenizers.git
 [submodule "kernels/optimized/third-party/eigen"]
 	path = kernels/optimized/third-party/eigen
 	url = https://gitlab.com/libeigen/eigen.git

backends/arm/_passes/TARGETS

@@ -6,6 +6,7 @@ python_library(
     deps = [
         "//executorch/backends/arm:tosa_quant_utils",
         "//executorch/backends/arm:tosa_utils",
+        "//executorch/backends/transforms:fuse_view_copy",
         "//executorch/backends/transforms:replace_scalar_with_tensor",
         "//executorch/backends/xnnpack/_passes:xnnpack_passes",
         "//executorch/exir:lib",

backends/arm/operator_support/tosa_supported_operators.py

@@ -112,6 +112,9 @@ def is_node_supported(
         supported = node.op == "call_function" and node.target in [
             exir_ops.edge.aten.abs.default,
             exir_ops.edge.aten.add.Tensor,
+            exir_ops.edge.aten.logical_and.default,
+            exir_ops.edge.aten.logical_or.default,
+            exir_ops.edge.aten.logical_xor.default,
             exir_ops.edge.aten.bitwise_and.Tensor,
             exir_ops.edge.aten.bitwise_or.Tensor,
             exir_ops.edge.aten.bitwise_xor.Tensor,
@@ -193,6 +196,9 @@ def is_node_supported(
                 exir_ops.edge.aten.bitwise_and.Tensor,
                 exir_ops.edge.aten.bitwise_or.Tensor,
                 exir_ops.edge.aten.bitwise_xor.Tensor,
+                exir_ops.edge.aten.logical_and.default,
+                exir_ops.edge.aten.logical_or.default,
+                exir_ops.edge.aten.logical_xor.default,
                 exir_ops.edge.aten.amax.default,
                 exir_ops.edge.aten.amin.default,
                 exir_ops.edge.aten.eq.Tensor,