xref: /aosp_15_r20/external/executorch/kernels/portable/cpu/op_add.cpp (revision 523fa7a60841cd1ecfb9cc4201f1ca8b03ed023a)
1 /*
2  * Copyright (c) Meta Platforms, Inc. and affiliates.
3  * All rights reserved.
4  *
5  * This source code is licensed under the BSD-style license found in the
6  * LICENSE file in the root directory of this source tree.
7  */
8 
9 #include <executorch/kernels/portable/cpu/scalar_utils.h>
10 #include <executorch/kernels/portable/cpu/util/elementwise_util.h>
11 #include <executorch/kernels/portable/cpu/util/kernel_ops_util.h>
12 #include <executorch/runtime/kernel/kernel_includes.h>
13 #include <executorch/runtime/platform/assert.h>
14 
15 namespace torch {
16 namespace executor {
17 namespace native {
18 
add_out(KernelRuntimeContext & ctx,const Tensor & a,const Tensor & b,const Scalar & alpha,Tensor & out)19 Tensor& add_out(
20     KernelRuntimeContext& ctx,
21     const Tensor& a,
22     const Tensor& b,
23     const Scalar& alpha,
24     Tensor& out) {
25   // Common Dtype
26   ScalarType common_type = promoteTypes(a.scalar_type(), b.scalar_type());
27 
28   // Check Common Dtype
29   ET_KERNEL_CHECK(
30       ctx,
31       (canCast(common_type, out.scalar_type()) &&
32        check_alpha_type(utils::get_scalar_dtype(alpha), common_type)),
33       InvalidArgument,
34       out);
35 
36   // Check Dim Order
37   ET_KERNEL_CHECK(
38       ctx, tensors_have_same_dim_order(a, b, out), InvalidArgument, out);
39 
40   // Resize
41   ET_KERNEL_CHECK(
42       ctx,
43       resize_to_broadcast_target_size(a, b, out) == Error::Ok,
44       InvalidArgument,
45       out);
46 
47   // Compute Dtype
48   ScalarType compute_type = utils::get_compute_type(common_type);
49 
50   // @lint-ignore CLANGTIDY facebook-hte-CArray
51   static constexpr const char op_name[] = "add.out";
52 
53   ET_SWITCH_REALB_TYPES(compute_type, ctx, op_name, CTYPE_COMPUTE, [&]() {
54     const CTYPE_COMPUTE val_alpha = utils::scalar_to<CTYPE_COMPUTE>(alpha);
55     utils::apply_bitensor_elementwise_fn<CTYPE_COMPUTE, op_name>(
56         [val_alpha](const CTYPE_COMPUTE val_a, const CTYPE_COMPUTE val_b) {
57           return val_a + val_alpha * val_b;
58         },
59         ctx,
60         a,
61         utils::SupportedTensorDtypes::REALHBBF16,
62         b,
63         utils::SupportedTensorDtypes::REALHBBF16,
64         out,
65         utils::SupportedTensorDtypes::REALHBBF16);
66   });
67 
68   return out;
69 }
70 
add_scalar_out(KernelRuntimeContext & ctx,const Tensor & a,const Scalar & b,const Scalar & alpha,Tensor & out)71 Tensor& add_scalar_out(
72     KernelRuntimeContext& ctx,
73     const Tensor& a,
74     const Scalar& b,
75     const Scalar& alpha,
76     Tensor& out) {
77   // Common Dtype
78   ScalarType common_type = utils::promote_type_with_scalar(a.scalar_type(), b);
79 
80   // Check Common Dtype
81   ET_KERNEL_CHECK(
82       ctx,
83       (common_type == out.scalar_type() &&
84        check_alpha_type(utils::get_scalar_dtype(alpha), common_type)),
85       InvalidArgument,
86       out);
87 
88   // Check Dim Order
89   ET_KERNEL_CHECK(
90       ctx, tensors_have_same_dim_order(a, out), InvalidArgument, out);
91 
92   // Resize
93   ET_KERNEL_CHECK(
94       ctx, resize_tensor(out, a.sizes()) == Error::Ok, InvalidArgument, out);
95 
96   // Compute Dtype
97   ScalarType compute_type = utils::get_compute_type(common_type);
98 
99   // @lint-ignore CLANGTIDY facebook-hte-CArray
100   static constexpr const char op_name[] = "add.Scalar_out";
101 
102   ET_SWITCH_REALB_TYPES(compute_type, ctx, op_name, CTYPE_COMPUTE, [&]() {
103     utils::apply_unitensor_elementwise_fn<CTYPE_COMPUTE, op_name>(
104         [b, alpha](const CTYPE_COMPUTE val_a) {
105           CTYPE_COMPUTE val_b = utils::scalar_to<CTYPE_COMPUTE>(b);
106           CTYPE_COMPUTE val_alpha = utils::scalar_to<CTYPE_COMPUTE>(alpha);
107           return val_a + val_alpha * val_b;
108         },
109         ctx,
110         a,
111         utils::SupportedTensorDtypes::REALHBBF16,
112         out,
113         utils::SupportedTensorDtypes::SAME_AS_COMMON);
114   });
115 
116   return out;
117 }
118 
119 } // namespace native
120 } // namespace executor
121 } // namespace torch
122