Update base for Update on "[Exutorch] Add broadcast support for le op"

For refactored hf repro requires this to support mask generation

Differential Revision: [D76456398](https://our.internmc.facebook.com/intern/diff/D76456398/)

[ghstack-poisoned]

Author: kimishpatel

.ci/scripts/test_llama.sh

@@ -156,8 +156,7 @@ cmake_install_executorch_libraries() {
         -DCMAKE_INSTALL_PREFIX=cmake-out \
         -DCMAKE_BUILD_TYPE="$CMAKE_BUILD_TYPE" \
         -DEXECUTORCH_BUILD_QNN="$QNN" \
-        -DQNN_SDK_ROOT="$QNN_SDK_ROOT" \
-        -Bcmake-out .
+        -DQNN_SDK_ROOT="$QNN_SDK_ROOT"
     cmake --build cmake-out -j9 --target install --config "$CMAKE_BUILD_TYPE"
 }
 

.ci/scripts/unittest-buck2.sh

@@ -25,7 +25,9 @@ BUILDABLE_KERNELS_PRIM_OPS_TARGETS=$(buck2 query //kernels/prim_ops/... | grep -
 # //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.
 for op in "build" "test"; do
-    buck2 $op $BUILDABLE_OPTIMIZED_OPS //kernels/portable/... \
+    buck2 $op $BUILDABLE_OPTIMIZED_OPS \
+          //examples/selective_build:select_all_dtype_selective_lib_portable_lib \
+          //kernels/portable/... \
           $BUILDABLE_KERNELS_PRIM_OPS_TARGETS //runtime/backend/... //runtime/core/... \
           //runtime/executor: //runtime/kernel/... //runtime/platform/...
 done

.lintrunner.toml

@@ -271,10 +271,6 @@ exclude_patterns = [
     'examples/**',
     'exir/verification/bindings.cpp',
     'extension/**',
-    # Uses properly-gated (ET_USE_PYTORCH_HEADERS) ATen include.
-    'kernels/portable/cpu/util/elementwise_util.h',
-    'kernels/portable/cpu/util/math_util.h',
-    'kernels/portable/cpu/util/vectorized_math.h',
     'kernels/optimized/**',
     'runtime/core/exec_aten/**',
     # Want to be able to keep c10 in sync with PyTorch core.

CMakeLists.txt

@@ -554,6 +554,10 @@ if(EXECUTORCH_BUILD_PTHREADPOOL AND EXECUTORCH_BUILD_CPUINFO)
 endif()
 
 if(EXECUTORCH_BUILD_PYBIND)
+
+  # Add codegen tools subdirectory for selective_build pybind module
+  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/codegen/tools)
+
   if(NOT EXECUTORCH_BUILD_EXTENSION_DATA_LOADER)
     add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/data_loader)
   endif()

backends/cadence/fusion_g3/operators/op_clamp.cpp

@@ -21,13 +21,13 @@
 #include <executorch/kernels/portable/cpu/util/math_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
 
-using ::executorch::aten::optional;
 using ::executorch::aten::Scalar;
 using ::executorch::aten::ScalarType;
 using ::executorch::aten::Tensor;
 using ::executorch::runtime::canCast;
 using ::executorch::runtime::Error;
 using ::executorch::runtime::KernelRuntimeContext;
+using std::optional;
 
 namespace cadence {
 namespace impl {

backends/cadence/fusion_g3/operators/op_dequantize.cpp

@@ -24,7 +24,7 @@ using ::executorch::runtime::Error;
 using ::executorch::runtime::KernelRuntimeContext;
 
 template <typename T>
-using optional = ::executorch::aten::optional<T>;
+using optional = std::optional<T>;
 /* ScalarType in Executorch do not have support for below data types.
  * So, creating a placeholder for these data types. Once, ScalarTypes is
  * updated to have support for below data types, these can be removed and
@@ -51,7 +51,7 @@ void check_dequantize_per_tensor_args(
     int64_t quant_min,
     int64_t quant_max,
     ScalarType dtype,
-    ::executorch::aten::optional<ScalarType>& out_dtype,
+    std::optional<ScalarType>& out_dtype,
     Tensor& out) {
   ET_CHECK_MSG(
       input.scalar_type() == ScalarType::Byte ||
@@ -93,7 +93,7 @@ Tensor& dequantize_impl(
     float* scale_data,
     int* zero_point_data,
     int* axis,
-    ::executorch::aten::optional<ScalarType> out_dtype) {
+    std::optional<ScalarType> out_dtype) {
   const ::executorch::aten::ArrayRef<Tensor::SizesType> input_size =
       input.sizes();
 
@@ -260,8 +260,8 @@ Tensor& dequantize_impl(
           }
         }
 
-        ::executorch::aten::optional<::executorch::aten::ArrayRef<int64_t>>
-            optional_dim_list{::executorch::aten::ArrayRef<int64_t>{
+        std::optional<::executorch::aten::ArrayRef<int64_t>> optional_dim_list{
+            ::executorch::aten::ArrayRef<int64_t>{
                 dims, size_t(input.dim() - 1)}};
 
 // Actual dequantization logic
@@ -466,8 +466,8 @@ Tensor& dequantize_impl(
           }
         }
 
-        ::executorch::aten::optional<::executorch::aten::ArrayRef<int64_t>>
-            optional_dim_list{::executorch::aten::ArrayRef<int64_t>{
+        std::optional<::executorch::aten::ArrayRef<int64_t>> optional_dim_list{
+            ::executorch::aten::ArrayRef<int64_t>{
                 dims, size_t(input.dim() - 1)}};
 
 // Actual dequantization logic
@@ -600,7 +600,7 @@ Tensor& dequantize_per_tensor_tensor_args_out(
     int64_t quant_min,
     int64_t quant_max,
     ScalarType dtype,
-    ::executorch::aten::optional<ScalarType> out_dtype,
+    std::optional<ScalarType> out_dtype,
     Tensor& out) {
 #ifdef OP_ARG_CHECK
   ET_CHECK_MSG(
@@ -639,12 +639,12 @@ Tensor& dequantize_per_channel_out(
     KernelRuntimeContext& context,
     const Tensor& input,
     const Tensor& scale,
-    const ::executorch::aten::optional<Tensor>& opt_zero_points,
+    const std::optional<Tensor>& opt_zero_points,
     int64_t axis,
     int64_t quant_min,
     int64_t quant_max,
     ScalarType dtype,
-    ::executorch::aten::optional<ScalarType> out_dtype,
+    std::optional<ScalarType> out_dtype,
     Tensor& out) {
   if (axis < 0) {
     axis += executorch::runtime::nonzero_dim(input);

backends/cadence/fusion_g3/operators/op_div.cpp

@@ -19,14 +19,14 @@
 #include <executorch/runtime/kernel/kernel_includes.h>
 #include <executorch/runtime/platform/assert.h>
 
-using ::executorch::aten::optional;
 using ::executorch::aten::Scalar;
 using ::executorch::aten::ScalarType;
-using ::executorch::aten::string_view;
 using ::executorch::aten::Tensor;
 using ::executorch::runtime::canCast;
 using ::executorch::runtime::Error;
 using ::executorch::runtime::KernelRuntimeContext;
+using std::optional;
+using std::string_view;
 
 namespace cadence {
 namespace impl {
@@ -686,4 +686,4 @@ Tensor& div_scalar_mode_out(
 } // namespace native
 } // namespace G3
 } // namespace impl
-} // namespace cadence
+} // namespace cadence

backends/cadence/fusion_g3/operators/op_mean.cpp

@@ -17,11 +17,11 @@
 #include <executorch/runtime/platform/assert.h>
 
 using ::executorch::aten::ArrayRef;
-using ::executorch::aten::optional;
 using ::executorch::aten::ScalarType;
 using ::executorch::aten::Tensor;
 using ::executorch::runtime::Error;
 using ::executorch::runtime::KernelRuntimeContext;
+using std::optional;
 
 namespace cadence {
 namespace impl {

backends/cadence/fusion_g3/operators/op_native_layer_norm.cpp

@@ -19,11 +19,11 @@
 #include <executorch/runtime/kernel/kernel_includes.h>
 
 using ::executorch::aten::IntArrayRef;
-using ::executorch::aten::optional;
 using ::executorch::aten::ScalarType;
 using ::executorch::aten::Tensor;
 using ::executorch::runtime::Error;
 using ::executorch::runtime::KernelRuntimeContext;
+using std::optional;
 
 namespace cadence {
 namespace impl {

backends/cadence/fusion_g3/operators/op_quantize.cpp

@@ -329,8 +329,8 @@ Tensor& quantize_impl(
           }
         }
 
-        ::executorch::aten::optional<::executorch::aten::ArrayRef<int64_t>>
-            optional_dim_list{::executorch::aten::ArrayRef<int64_t>{
+        std::optional<::executorch::aten::ArrayRef<int64_t>> optional_dim_list{
+            ::executorch::aten::ArrayRef<int64_t>{
                 dims, size_t(input.dim() - 1)}};
 
 // Actual quantization logic
@@ -534,8 +534,8 @@ Tensor& quantize_impl(
           }
         }
 
-        ::executorch::aten::optional<::executorch::aten::ArrayRef<int64_t>>
-            optional_dim_list{::executorch::aten::ArrayRef<int64_t>{
+        std::optional<::executorch::aten::ArrayRef<int64_t>> optional_dim_list{
+            ::executorch::aten::ArrayRef<int64_t>{
                 dims, size_t(input.dim() - 1)}};
 
 // Actual quantization logic