1 | /******************************************************************************* |
2 | * Copyright 2021-2022 Intel Corporation |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
5 | * you may not use this file except in compliance with the License. |
6 | * You may obtain a copy of the License at |
7 | * |
8 | * http://www.apache.org/licenses/LICENSE-2.0 |
9 | * |
10 | * Unless required by applicable law or agreed to in writing, software |
11 | * distributed under the License is distributed on an "AS IS" BASIS, |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
13 | * See the License for the specific language governing permissions and |
14 | * limitations under the License. |
15 | *******************************************************************************/ |
16 | |
17 | #ifndef CPU_REF_INNER_PRODUCT_INT8_HPP |
18 | #define CPU_REF_INNER_PRODUCT_INT8_HPP |
19 | |
20 | #include <assert.h> |
21 | |
22 | #include "common/c_types_map.hpp" |
23 | #include "common/primitive.hpp" |
24 | #include "common/type_helpers.hpp" |
25 | #include "common/utils.hpp" |
26 | |
27 | #include "cpu/cpu_inner_product_pd.hpp" |
28 | #include "cpu/primitive_attr_postops.hpp" |
29 | |
30 | namespace dnnl { |
31 | namespace impl { |
32 | namespace cpu { |
33 | |
34 | struct ref_inner_product_int8_fwd_t : public primitive_t { |
35 | struct pd_t : public cpu_inner_product_fwd_pd_t { |
36 | using cpu_inner_product_fwd_pd_t::cpu_inner_product_fwd_pd_t; |
37 | |
38 | DECLARE_COMMON_PD_T("ref:any" , ref_inner_product_int8_fwd_t); |
39 | |
40 | status_t init(engine_t *engine) { |
41 | using namespace data_type; |
42 | using smask_t = primitive_attr_t::skip_mask_t; |
43 | const auto src_type = src_md(0)->data_type; |
44 | const auto wei_type = weights_md(0)->data_type; |
45 | const auto bia_type = weights_md(1)->data_type; |
46 | const auto dst_type = dst_md(0)->data_type; |
47 | |
48 | const bool allow_all_tags = true; // ref should support all tags |
49 | |
50 | bool ok = is_fwd() && utils::one_of(src_type, s8, u8) |
51 | && wei_type == s8 |
52 | && IMPLICATION(with_bias(), |
53 | utils::one_of(bia_type, f32, bf16, s32, s8, u8)) |
54 | && utils::one_of(dst_type, f32, bf16, s32, s8, u8) |
55 | && IMPLICATION(with_bias(), |
56 | platform::has_data_type_support(bia_type)) |
57 | && platform::has_data_type_support(dst_type) |
58 | && set_default_params(allow_all_tags) == status::success |
59 | && attr()->has_default_values(smask_t::scales_runtime |
60 | | smask_t::post_ops | smask_t::sum_dt) |
61 | && attr()->post_ops_.check_sum_consistent_dt(dst_type) |
62 | && arg_scales_mask_ok() |
63 | && attr_.set_default_formats(dst_md(0)) == status::success; |
64 | return ok ? status::success : status::unimplemented; |
65 | } |
66 | |
67 | private: |
68 | bool arg_scales_mask_ok() const { |
69 | const std::vector<int> supported_args |
70 | = {DNNL_ARG_SRC, DNNL_ARG_WEIGHTS, DNNL_ARG_DST}; |
71 | bool ok = attr()->scales_.has_default_values(supported_args); |
72 | |
73 | for (auto arg : supported_args) { |
74 | int mask = attr()->scales_.get(arg).mask_; |
75 | if (arg == DNNL_ARG_WEIGHTS) |
76 | ok = ok && (mask == 0 || mask == (1 << 0)); |
77 | else |
78 | ok = ok && (mask == 0); |
79 | } |
80 | return ok; |
81 | } |
82 | }; |
83 | |
84 | ref_inner_product_int8_fwd_t(const pd_t *apd) : primitive_t(apd) {} |
85 | |
86 | status_t init(engine_t *engine) override { |
87 | ref_post_ops |
88 | = utils::make_unique<ref_post_ops_t>(pd()->attr()->post_ops_); |
89 | if (!ref_post_ops) return status::out_of_memory; |
90 | return status::success; |
91 | } |
92 | |
93 | status_t execute(const exec_ctx_t &ctx) const override { |
94 | return execute_forward(ctx); |
95 | } |
96 | |
97 | private: |
98 | status_t execute_forward(const exec_ctx_t &ctx) const; |
99 | const pd_t *pd() const { return (const pd_t *)primitive_t::pd().get(); } |
100 | std::unique_ptr<ref_post_ops_t> ref_post_ops; |
101 | }; |
102 | |
103 | } // namespace cpu |
104 | } // namespace impl |
105 | } // namespace dnnl |
106 | |
107 | #endif |
108 | |
109 | // vim: et ts=4 sw=4 cindent cino+=l0,\:4,N-s |
110 | |