|
5 | 5 | namespace torch {
|
6 | 6 | namespace executor {
|
7 | 7 | namespace native {
|
8 |
| -using Tensor = exec_aten::Tensor; |
9 |
| -using ScalarType = exec_aten::ScalarType; |
10 |
| -using Scalar = exec_aten::Scalar; |
11 |
| -namespace { |
12 |
| - |
13 |
| -template <typename CTYPE_A, typename CTYPE_B, typename CTYPE_OUT> |
14 |
| -void where_tensors_impl( |
15 |
| - const Tensor& condition, |
16 |
| - const Tensor& a, |
17 |
| - const Tensor& b, |
18 |
| - Tensor& out) { |
19 |
| - apply_ternary_elementwise_fn<CTYPE_A, CTYPE_B, bool, CTYPE_OUT>( |
20 |
| - [](const CTYPE_A val_a, const CTYPE_B val_b, const bool val_c) { |
21 |
| - CTYPE_OUT a_casted = static_cast<CTYPE_OUT>(val_a); |
22 |
| - CTYPE_OUT b_casted = static_cast<CTYPE_OUT>(val_b); |
23 |
| - |
24 |
| - return val_c ? a_casted : b_casted; |
25 |
| - }, |
26 |
| - a, |
27 |
| - b, |
28 |
| - condition, |
29 |
| - out); |
30 |
| -} |
31 |
| - |
32 |
| -template <typename CTYPE_A, typename CTYPE_B> |
33 |
| -void where_tensors_switch_out( |
34 |
| - const Tensor& condition, |
35 |
| - const Tensor& a, |
36 |
| - const Tensor& b, |
37 |
| - Tensor& out) { |
38 |
| -#define WHERE_TENSORS_SWITCH_OUT_CASE(ctype, dtype) \ |
39 |
| - case ScalarType::dtype: \ |
40 |
| - where_tensors_impl<CTYPE_A, CTYPE_B, ctype>(condition, a, b, out); \ |
41 |
| - break; |
42 |
| - |
43 |
| - switch (out.scalar_type()) { |
44 |
| - ET_FORALL_REAL_TYPES_AND(Bool, WHERE_TENSORS_SWITCH_OUT_CASE) |
45 |
| - default: |
46 |
| - ET_CHECK_MSG(false, "Unhandled dtype %hhd for out", out.scalar_type()); |
47 |
| - } |
48 |
| - |
49 |
| -#undef WHERE_TENSORS_SWITCH_OUT_CASE |
50 |
| -} |
51 |
| - |
52 |
| -template <typename CTYPE_A> |
53 |
| -void where_tensors_switch_b( |
54 |
| - const Tensor& condition, |
55 |
| - const Tensor& a, |
56 |
| - const Tensor& b, |
57 |
| - Tensor& out) { |
58 |
| -#define WHERE_TENSORS_SWITCH_B_CASE(ctype, dtype) \ |
59 |
| - case ScalarType::dtype: \ |
60 |
| - where_tensors_switch_out<CTYPE_A, ctype>(condition, a, b, out); \ |
61 |
| - break; |
62 |
| - |
63 |
| - switch (b.scalar_type()) { |
64 |
| - ET_FORALL_REAL_TYPES_AND(Bool, WHERE_TENSORS_SWITCH_B_CASE) |
65 |
| - default: |
66 |
| - ET_CHECK_MSG(false, "Unhandled dtype %hhd for b", b.scalar_type()); |
67 |
| - } |
68 |
| - |
69 |
| -#undef WHERE_TENSORS_SWITCH_B_CASE |
70 |
| -} |
71 |
| - |
72 |
| -void where_tensors_switch_a( |
73 |
| - const Tensor& condition, |
74 |
| - const Tensor& a, |
75 |
| - const Tensor& b, |
76 |
| - Tensor& out) { |
77 |
| -#define WHERE_TENSORS_SWITCH_A_CASE(ctype, dtype) \ |
78 |
| - case ScalarType::dtype: \ |
79 |
| - where_tensors_switch_b<ctype>(condition, a, b, out); \ |
80 |
| - break; |
81 |
| - |
82 |
| - switch (a.scalar_type()) { |
83 |
| - ET_FORALL_REAL_TYPES_AND(Bool, WHERE_TENSORS_SWITCH_A_CASE) |
84 |
| - default: |
85 |
| - ET_CHECK_MSG(false, "Unhandled dtype %hhd for a", a.scalar_type()); |
86 |
| - } |
87 |
| - |
88 |
| -#undef WHERE_TENSORS_SWITCH_A_CASE |
89 |
| -} |
90 |
| - |
91 |
| -void check_input_dtypes( |
92 |
| - const Tensor& condition, |
93 |
| - const Tensor& a, |
94 |
| - const Tensor& b, |
95 |
| - Tensor& out) { |
96 |
| - ET_CHECK_MSG( |
97 |
| - condition.scalar_type() == ScalarType::Bool, |
98 |
| - "Condition tensor must be boolean type"); |
99 |
| -} |
100 |
| - |
101 |
| -} // namespace |
102 | 8 |
|
103 | 9 | Tensor& where_out(
|
104 |
| - RuntimeContext& context, |
105 |
| - const Tensor& condition, |
| 10 | + RuntimeContext& ctx, |
| 11 | + const Tensor& cond, |
106 | 12 | const Tensor& a,
|
107 | 13 | const Tensor& b,
|
108 | 14 | Tensor& out) {
|
109 |
| - (void)context; |
| 15 | + (void)ctx; |
110 | 16 |
|
111 |
| - // Determine output size and resize for dynamic shapes |
112 |
| - resize_to_broadcast_target_size(a, b, condition, out); |
| 17 | + ScalarType cond_type = cond.scalar_type(); |
| 18 | + ScalarType a_type = a.scalar_type(); |
| 19 | + ScalarType b_type = b.scalar_type(); |
| 20 | + ScalarType common_type = promoteTypes(a_type, b_type); |
| 21 | + ScalarType out_type = out.scalar_type(); |
113 | 22 |
|
114 |
| - // Check arguments |
115 |
| - check_input_dtypes(condition, a, b, out); |
| 23 | + ET_CHECK(common_type == out_type); |
116 | 24 |
|
117 |
| - where_tensors_switch_a(condition, a, b, out); |
| 25 | + // Determine output size and resize for dynamic shapes |
| 26 | + resize_to_broadcast_target_size(a, b, cond, out); |
| 27 | + |
| 28 | + ET_SWITCH_TWO_TYPES(Bool, Byte, cond_type, ctx, "where", CTYPE_COND, [&]() { |
| 29 | + ET_SWITCH_REAL_TYPES_AND(Bool, a_type, ctx, "where", CTYPE_A, [&]() { |
| 30 | + ET_SWITCH_REAL_TYPES_AND(Bool, b_type, ctx, "where", CTYPE_B, [&]() { |
| 31 | + ET_SWITCH_REAL_TYPES_AND( |
| 32 | + Bool, out_type, ctx, "where", CTYPE_OUT, [&]() { |
| 33 | + apply_ternary_elementwise_fn< |
| 34 | + CTYPE_A, |
| 35 | + CTYPE_B, |
| 36 | + CTYPE_COND, |
| 37 | + CTYPE_OUT>( |
| 38 | + [](const CTYPE_A val_a, |
| 39 | + const CTYPE_B val_b, |
| 40 | + const CTYPE_COND val_c) { |
| 41 | + CTYPE_OUT a_casted = static_cast<CTYPE_OUT>(val_a); |
| 42 | + CTYPE_OUT b_casted = static_cast<CTYPE_OUT>(val_b); |
| 43 | + return val_c ? a_casted : b_casted; |
| 44 | + }, |
| 45 | + a, |
| 46 | + b, |
| 47 | + cond, |
| 48 | + out); |
| 49 | + }); |
| 50 | + }); |
| 51 | + }); |
| 52 | + }); |
118 | 53 |
|
119 | 54 | return out;
|
120 | 55 | }
|
|
0 commit comments