Update on "[ET-VK] Adding push constant and ubo verison of select and slice ops to improve memory and performance."

Adding push constant and ubo verison of select and slice ops to improve memory and performance.
*   Updated `transfer_buffer.yaml` and `transfer_texture.yaml` to include `UBO_PARAMS` parameter and generate variants for `select` and `slice` ops with UBO parameters.
*   Updated `transfer.glsl` to generate ubo and push constant versions of `select` and `slice` ops with UBO parameters.

Differential Revision: [D78095262](https://our.internmc.facebook.com/intern/diff/D78095262/)

[ghstack-poisoned]

Author: trivedivivek

.Package.swift/kernels_custom/dummy.swift renamed to .Package.swift/kernels_llm/dummy.swift

@@ -0,0 +1,13 @@
+
+set -ex
+
+install_emscripten() {
+    git clone https://github.com/emscripten-core/emsdk.git
+    pushd emsdk || return
+    ./emsdk install 4.0.10
+    ./emsdk activate 4.0.10
+    source ./emsdk_env.sh
+    popd || return
+}
+
+install_emscripten

.Package.swift/kernels_custom_debug/dummy.swift renamed to .Package.swift/kernels_llm_debug/dummy.swift

@@ -35,15 +35,15 @@ cmake -DPYTHON_EXECUTABLE=python \
     -DEXECUTORCH_BUILD_XNNPACK=OFF \
     -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
     -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
-    -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON \
+    -DEXECUTORCH_BUILD_KERNELS_LLM=ON \
     -Bcmake-out .
 cmake --build cmake-out -j16 --target install --config Release
 
 # Install llama runner with torchao
 cmake -DPYTHON_EXECUTABLE=python \
     -DBUILD_TESTING=OFF \
     -DCMAKE_BUILD_TYPE=Release \
-    -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON \
+    -DEXECUTORCH_BUILD_KERNELS_LLM=ON \
     -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
     -DEXECUTORCH_BUILD_XNNPACK=OFF \
     -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \

.ci/scripts/setup-emscripten.sh

@@ -39,7 +39,7 @@ EXECUTORCH_COMMON_CMAKE_ARGS="                      \
         -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
         -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
         -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON      \
-        -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON        \
+        -DEXECUTORCH_BUILD_KERNELS_LLM=ON        \
         -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON     \
         -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON     \
         -DEXECUTORCH_BUILD_XNNPACK=ON               \
@@ -69,7 +69,7 @@ LLAVA_COMMON_CMAKE_ARGS="                        \
         -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" \
         -DCMAKE_INSTALL_PREFIX=${BUILD_DIR}      \
         -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}   \
-        -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON     \
+        -DEXECUTORCH_BUILD_KERNELS_LLM=ON     \
         -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON  \
         -DEXECUTORCH_BUILD_XNNPACK=ON"
 

.ci/scripts/test_llama_torchao_lowbit.sh

@@ -33,7 +33,7 @@ cmake_install_executorch_libraries() {
       -DEXECUTORCH_BUILD_XNNPACK=ON \
       -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
       -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
-      -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON \
+      -DEXECUTORCH_BUILD_KERNELS_LLM=ON \
       -B${BUILD_DIR} .
 
   cmake --build ${BUILD_DIR} -j${NPROC} --target install --config ${BUILD_TYPE}
@@ -43,7 +43,7 @@ cmake_build_phi_3_mini() {
   cmake -DPYTHON_EXECUTABLE=$PYTHON_EXECUTABLE \
       -DCMAKE_INSTALL_PREFIX=${BUILD_DIR} \
       -DCMAKE_BUILD_TYPE=${BUILD_TYPE} \
-      -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON \
+      -DEXECUTORCH_BUILD_KERNELS_LLM=ON \
       -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
       -DEXECUTORCH_BUILD_XNNPACK=ON \
       -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \

.ci/scripts/test_llava.sh

@@ -152,7 +152,7 @@ jobs:
           "backend_coreml"
           "backend_mps"
           "backend_xnnpack"
-          "kernels_custom"
+          "kernels_llm"
           "kernels_optimized"
           "kernels_quantized"
           "threadpool"

.ci/scripts/test_phi_3_mini.sh

@@ -734,3 +734,31 @@ jobs:
 
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-openvino.sh
         PYTHON_EXECUTABLE=python bash .ci/scripts/test_openvino.sh
+
+  test-build-wasm-linux:
+    name: test-build-wasm-linux
+    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
+    permissions:
+      id-token: write
+      contents: read
+    strategy:
+      fail-fast: false
+    with:
+      runner: linux.2xlarge
+      docker-image: executorch-ubuntu-22.04-clang12
+      submodules: 'recursive'
+      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      timeout: 90
+      script: |
+        # The generic Linux job chooses to use base env, not the one setup by the image
+        CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
+        conda activate "${CONDA_ENV}"
+
+        BUILD_TOOL="cmake"
+        PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh --build-tool "${BUILD_TOOL}"
+
+        # Install Node.js and Emscripten
+        source .ci/scripts/setup-emscripten.sh
+
+        # Test selective build
+        PYTHON_EXECUTABLE=python bash examples/wasm/test_build_wasm.sh

.github/workflows/apple.yml

@@ -645,7 +645,7 @@ jobs:
           -DEXECUTORCH_BUILD_XNNPACK=ON \
           -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
           -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
-          -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON \
+          -DEXECUTORCH_BUILD_KERNELS_LLM=ON \
           -DEXECUTORCH_BUILD_DEVTOOLS=ON \
           -DEXECUTORCH_ENABLE_EVENT_TRACER=ON \
           -Bcmake-out .

.github/workflows/pull.yml

@@ -26,6 +26,7 @@ pip-out/
 *.model
 tokenizer.json
 *.pte
+*.ptd
 !test_bpe_tokenizer.bin
 !test_tiktoken_tokenizer.model