• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/broadcast_util.h>
11 #include <executorch/runtime/kernel/kernel_includes.h>
12 #include <executorch/runtime/platform/assert.h>
13 
14 namespace torch {
15 namespace executor {
16 namespace native {
17 
add_out(KernelRuntimeContext & ctx,const Tensor & a,const Tensor & b,const Scalar & alpha,Tensor & out)18 Tensor& add_out(
19     KernelRuntimeContext& ctx,
20     const Tensor& a,
21     const Tensor& b,
22     const Scalar& alpha,
23     Tensor& out) {
24   (void)ctx;
25 
26   ScalarType a_type = a.scalar_type();
27   ScalarType b_type = b.scalar_type();
28   ScalarType common_type = promoteTypes(a_type, b_type);
29   ScalarType out_type = out.scalar_type();
30 
31   ET_CHECK_MSG(a_type == ScalarType::Float, "Input tensor not a float.\n");
32   ET_CHECK_MSG(b_type == ScalarType::Float, "Input tensor not a float.\n");
33   ET_CHECK_MSG(out_type == ScalarType::Float, "Output tensor not a float.\n");
34 
35   ET_CHECK(canCast(common_type, out_type));
36 
37   using CTYPE_A = float;
38   using CTYPE_B = float;
39   using CTYPE_IN = float;
40   using CTYPE_OUT = float;
41   CTYPE_IN alpha_val;
42   ET_EXTRACT_SCALAR(alpha, alpha_val);
43 
44   apply_binary_elementwise_fn<CTYPE_A, CTYPE_B, CTYPE_OUT>(
45       [alpha_val](const CTYPE_A val_a, const CTYPE_B val_b) {
46         CTYPE_IN a_casted = static_cast<CTYPE_IN>(val_a);
47         CTYPE_IN b_casted = static_cast<CTYPE_IN>(val_b);
48         CTYPE_IN value = a_casted + alpha_val * b_casted;
49 
50         return static_cast<CTYPE_OUT>(value);
51       },
52       a,
53       b,
54       out);
55 
56   return out;
57 }
58 
59 } // namespace native
60 } // namespace executor
61 } // namespace torch
62