Revert vectorization stack (pytorch#11536)

### Summary
These broke some internal tests. @swolchok said to just revert and he
would try again

### Test plan
ci

Author: JacobSzwejbka

.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.

kernels/portable/CMakeLists.txt

@@ -69,15 +69,8 @@ if(EXECUTORCH_BUILD_PTHREADPOOL AND EXECUTORCH_BUILD_KERNELS_OPTIMIZED)
   target_compile_options(optimized_portable_kernels PUBLIC ${_common_compile_options})
   target_include_directories(optimized_portable_kernels PRIVATE ${TORCH_INCLUDE_DIRS})
   target_compile_definitions(optimized_portable_kernels PRIVATE ET_USE_PYTORCH_HEADERS)
-  gen_selected_ops(LIB_NAME "optimized_portable_ops_lib" OPS_SCHEMA_YAML "${_yaml}")
-  generate_bindings_for_kernels(
-    LIB_NAME "optimized_portable_ops_lib" FUNCTIONS_YAML "${_yaml}"
-  )
-  gen_operators_lib(
-    LIB_NAME "optimized_portable_ops_lib" KERNEL_LIBS optimized_portable_kernels DEPS executorch_core
-  )
   install(
-    TARGETS optimized_portable_kernels optimized_portable_ops_lib
+    TARGETS optimized_portable_kernels
     DESTINATION lib
   )
 endif()

kernels/portable/cpu/op_acos.cpp

@@ -15,9 +15,8 @@ namespace executor {
 namespace native {
 
 Tensor& acos_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  static constexpr const char op_name[] = "acos.out";
-  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
-      [](auto x) { return executorch::math::acos(x); }, ctx, in, out);
+  return internal::unary_ufunc_realhbbf16_to_floathbf16(
+      std::acos, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_acosh.cpp

@@ -15,9 +15,8 @@ namespace executor {
 namespace native {
 
 Tensor& acosh_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  static constexpr const char op_name[] = "acosh.out";
-  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
-      [](auto x) { return executorch::math::acosh(x); }, ctx, in, out);
+  return internal::unary_ufunc_realhbbf16_to_floathbf16(
+      std::acosh, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_add.cpp

@@ -102,18 +102,14 @@ Tensor& add_scalar_out(
   static constexpr const char op_name[] = "add.Scalar_out";
 
   ET_SWITCH_REALB_TYPES(compute_type, ctx, op_name, CTYPE_COMPUTE, [&]() {
-    CTYPE_COMPUTE val_b = utils::scalar_to<CTYPE_COMPUTE>(b);
-    CTYPE_COMPUTE val_alpha = utils::scalar_to<CTYPE_COMPUTE>(alpha);
-    auto val_alpha_times_b = val_alpha * val_b;
     utils::apply_unitensor_elementwise_fn<
         CTYPE_COMPUTE,
         op_name,
         utils::SupportedTensorDtypes::SAME_AS_COMMON>(
-        [val_alpha_times_b](const auto val_a) {
-          // Cast here supports vectorization; either it does nothing
-          // or it casts from CTYPE_COMPUTE to
-          // Vectorized<CTYPE_COMPUTE>.
-          return val_a + decltype(val_a)(val_alpha_times_b);
+        [b, alpha](const auto val_a) {
+          CTYPE_COMPUTE val_b = utils::scalar_to<CTYPE_COMPUTE>(b);
+          CTYPE_COMPUTE val_alpha = utils::scalar_to<CTYPE_COMPUTE>(alpha);
+          return val_a + val_alpha * val_b;
         },
         ctx,
         a,

kernels/portable/cpu/op_asin.cpp

@@ -15,9 +15,8 @@ namespace executor {
 namespace native {
 
 Tensor& asin_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  static constexpr const char op_name[] = "asin.out";
-  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
-      [](auto x) { return executorch::math::asin(x); }, ctx, in, out);
+  return internal::unary_ufunc_realhbbf16_to_floathbf16(
+      std::asin, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_asinh.cpp

@@ -15,9 +15,8 @@ namespace executor {
 namespace native {
 
 Tensor& asinh_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  static constexpr const char op_name[] = "asinh.out";
-  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
-      [](auto x) { return executorch::math::asinh(x); }, ctx, in, out);
+  return internal::unary_ufunc_realhbbf16_to_floathbf16(
+      std::asinh, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_atan.cpp

@@ -15,9 +15,8 @@ namespace executor {
 namespace native {
 
 Tensor& atan_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  static constexpr const char op_name[] = "atan.out";
-  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
-      [](auto x) { return executorch::math::atan(x); }, ctx, in, out);
+  return internal::unary_ufunc_realhbbf16_to_floathbf16(
+      std::atan, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_atan2.cpp

@@ -60,7 +60,7 @@ Tensor& atan2_out(
         op_name,
         utils::SupportedTensorDtypes::FLOATHBF16>(
         [](const auto val_a, const auto val_b) {
-          return executorch::math::atan2(val_a, val_b);
+          return std::atan2(val_a, val_b);
         },
         ctx,
         a,

kernels/portable/cpu/op_atanh.cpp

@@ -15,9 +15,8 @@ namespace executor {
 namespace native {
 
 Tensor& atanh_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  static constexpr const char op_name[] = "atanh.out";
-  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
-      [](auto x) { return executorch::math::atanh(x); }, ctx, in, out);
+  return internal::unary_ufunc_realhbbf16_to_floathbf16(
+      std::atanh, ctx, in, out);
 }
 
 } // namespace native