Implement unary_ufunc functions using elementwise_util

kernels/portable/cpu/op_acos.cpp

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

kernels/portable/cpu/op_acosh.cpp

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

kernels/portable/cpu/op_asin.cpp

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

kernels/portable/cpu/op_asinh.cpp

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

kernels/portable/cpu/op_atan.cpp

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

kernels/portable/cpu/op_atanh.cpp

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

kernels/portable/cpu/op_ceil.cpp

@@ -17,7 +17,9 @@ namespace native {
 using executorch::aten::Tensor;
 
 Tensor& ceil_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  return internal::unary_ufunc_realh(std::ceil, ctx, in, out);
+  static constexpr const char op_name[] = "ceil.out";
+  return internal::unary_ufunc_realh<op_name>(
+      [](auto x) { return std::ceil(x); }, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_cos.cpp

@@ -15,7 +15,9 @@ namespace executor {
 namespace native {
 
 Tensor& cos_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  return internal::unary_ufunc_realhbbf16_to_floathbf16(std::cos, ctx, in, out);
+  static constexpr const char op_name[] = "cos.out";
+  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
+      [](auto x) { return std::cos(x); }, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_cosh.cpp

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

kernels/portable/cpu/op_erf.cpp

@@ -15,7 +15,9 @@ namespace executor {
 namespace native {
 
 Tensor& erf_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  return internal::unary_ufunc_realhbbf16_to_floathbf16(std::erf, ctx, in, out);
+  static constexpr const char op_name[] = "erf.out";
+  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
+      [](auto x) { return std::erf(x); }, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_exp.cpp

@@ -15,7 +15,9 @@ namespace executor {
 namespace native {
 
 Tensor& exp_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  return internal::unary_ufunc_realhbbf16_to_floathbf16(std::exp, ctx, in, out);
+  static constexpr const char op_name[] = "exp.out";
+  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
+      [](auto x) { return std::exp(x); }, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_expm1.cpp

@@ -15,8 +15,9 @@ namespace executor {
 namespace native {
 
 Tensor& expm1_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  return internal::unary_ufunc_realhbbf16_to_floathbf16(
-      std::expm1, ctx, in, out);
+  static constexpr const char op_name[] = "expm1.out";
+  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
+      [](auto x) { return std::expm1(x); }, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_floor.cpp

@@ -17,7 +17,9 @@ namespace native {
 using executorch::aten::Tensor;
 
 Tensor& floor_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  return internal::unary_ufunc_realh(std::floor, ctx, in, out);
+  static constexpr const char op_name[] = "floor.out";
+  return internal::unary_ufunc_realh<op_name>(
+      [](auto x) { return std::floor(x); }, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_isinf.cpp

@@ -17,8 +17,9 @@ namespace native {
 Tensor& isinf_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
   // Lambda is syntactic sugar needed to workaround compilation on some older
   // non-compatible distros where isnan is returning int rather than bool
-  return internal::unary_ufunc_realhb_to_bool(
-      [](double x) -> bool { return std::isinf(x); }, ctx, in, out);
+  static constexpr const char op_name[] = "isinf.out";
+  return internal::unary_ufunc_realhb_to_bool<op_name>(
+      [](auto x) -> bool { return std::isinf(x); }, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_isnan.cpp

@@ -17,8 +17,9 @@ namespace native {
 Tensor& isnan_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
   // Lambda is syntactic sugar needed to workaround compilation on some older
   // non-compatible distros where isnan is returning int rather than bool
-  return internal::unary_ufunc_realhb_to_bool(
-      [](double x) -> bool { return std::isnan(x); }, ctx, in, out);
+  static constexpr const char op_name[] = "isnan.out";
+  return internal::unary_ufunc_realhb_to_bool<op_name>(
+      [](auto x) -> bool { return std::isnan(x); }, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_log.cpp

@@ -15,7 +15,9 @@ namespace executor {
 namespace native {
 
 Tensor& log_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  return internal::unary_ufunc_realhbbf16_to_floathbf16(std::log, ctx, in, out);
+  static constexpr const char op_name[] = "log.out";
+  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
+      [](auto x) { return std::log(x); }, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_log10.cpp

@@ -15,8 +15,9 @@ namespace executor {
 namespace native {
 
 Tensor& log10_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  return internal::unary_ufunc_realhbbf16_to_floathbf16(
-      std::log10, ctx, in, out);
+  static constexpr const char op_name[] = "log10.out";
+  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
+      [](auto x) { return std::log10(x); }, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_log1p.cpp

@@ -15,8 +15,9 @@ namespace executor {
 namespace native {
 
 Tensor& log1p_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  return internal::unary_ufunc_realhbbf16_to_floathbf16(
-      std::log1p, ctx, in, out);
+  static constexpr const char op_name[] = "log1p.out";
+  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
+      [](auto x) { return std::log1p(x); }, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_log2.cpp

@@ -15,8 +15,9 @@ namespace executor {
 namespace native {
 
 Tensor& log2_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  return internal::unary_ufunc_realhbbf16_to_floathbf16(
-      std::log2, ctx, in, out);
+  static constexpr const char op_name[] = "log2.out";
+  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
+      [](auto x) { return std::log2(x); }, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_reciprocal.cpp

@@ -14,16 +14,27 @@ namespace executor {
 namespace native {
 namespace {
 
+float reciprocal(float x) {
+  return 1.0f / x;
+}
+
 double reciprocal(double x) {
   return 1.0 / x;
 }
 
+template <
+    typename Integer,
+    std::enable_if_t<std::is_integral_v<Integer>, bool> = true>
+double reciprocal(Integer x) {
+  return reciprocal((double)x);
+}
 } // namespace
 
 Tensor&
 reciprocal_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  return internal::unary_ufunc_realhbbf16_to_floathbf16(
-      reciprocal, ctx, in, out);
+  static constexpr const char op_name[] = "reciprocal.out";
+  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
+      [](auto x) { return reciprocal(x); }, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_rsqrt.cpp

@@ -12,16 +12,17 @@
 namespace torch {
 namespace executor {
 namespace native {
-namespace {
-
-double rsqrt(double x) {
-  return 1.0 / std::sqrt(x);
-}
-
-} // namespace
 
 Tensor& rsqrt_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  return internal::unary_ufunc_realhbbf16_to_floathbf16(rsqrt, ctx, in, out);
+  static constexpr const char op_name[] = "rsqrt.out";
+  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
+      [](auto x) {
+        auto result = std::sqrt(x);
+        return static_cast<decltype(result)>(1) / result;
+      },
+      ctx,
+      in,
+      out);
 }
 
 } // namespace native

kernels/portable/cpu/op_sin.cpp

@@ -15,7 +15,9 @@ namespace executor {
 namespace native {
 
 Tensor& sin_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  return internal::unary_ufunc_realhbbf16_to_floathbf16(std::sin, ctx, in, out);
+  static constexpr const char op_name[] = "sin.out";
+  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
+      [](auto x) { return std::sin(x); }, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_sinh.cpp

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

kernels/portable/cpu/op_sqrt.cpp

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

kernels/portable/cpu/op_tan.cpp

@@ -15,7 +15,9 @@ namespace executor {
 namespace native {
 
 Tensor& tan_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  return internal::unary_ufunc_realhbbf16_to_floathbf16(std::tan, ctx, in, out);
+  static constexpr const char op_name[] = "tan.out";
+  return internal::unary_ufunc_realhbbf16_to_floathbf16<op_name>(
+      [](auto x) { return std::tan(x); }, ctx, in, out);
 }
 
 } // namespace native

kernels/portable/cpu/op_tanh.cpp

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

kernels/portable/cpu/op_trunc.cpp

@@ -15,7 +15,9 @@ namespace executor {
 namespace native {
 
 Tensor& trunc_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
-  return internal::unary_ufunc_realh(std::trunc, ctx, in, out);
+  static constexpr const char op_name[] = "trunc.out";
+  return internal::unary_ufunc_realh<op_name>(
+      [](auto x) { return std::trunc(x); }, ctx, in, out);
 }
 
 } // namespace native