onednn-src 0.1.13

Source of oneAPI Deep Neural Network Library (oneDNN)
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168

/*******************************************************************************
* Copyright 2024 Intel Corporation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
*     http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*******************************************************************************/

#include "gpu/intel/jit/ir/tensor_config.hpp"

namespace dnnl {
namespace impl {
namespace gpu {
namespace intel {
namespace jit {

void init_extra_tensors(const zero_points_config_t &zp_cfg,
        const primitive_attr_t &attr, const memory_desc_t *zp_src,
        const memory_desc_t &dst_md, dim_t ic, dim_t oc,
        tensor_config_t &tensor_cfg) {
    if (!attr.rounding_mode_.has_default_values()) {
        layout_t sround_seed_layout(dsl::type_t::u32(), std::vector<dim_t> {1});
        tensor_cfg.add_tensor("sround_seed", DNNL_ARG_ATTR_ROUNDING_SEED,
                /*is_input=*/true, /*is_output=*/false, sround_seed_layout);
    }
    if (attr.scales_.get(DNNL_ARG_DST).is_mx()) {
        layout_t dst_scales_layout(dsl::type_t::u32(), std::vector<dim_t> {1});
        tensor_cfg.add_tensor("dst_scales", DNNL_ARG_ATTR_SCALES | DNNL_ARG_DST,
                /*is_input=*/false, /*is_output=*/true, dst_scales_layout);
    }
    auto add_zp_buffer = [&](const std::string &name, dsl::type_t type,
                                 int arg_id, dim_t size) {
        layout_t zp_layout(type, std::vector<dim_t> {size});
        tensor_cfg.add_tensor(name, DNNL_ARG_ATTR_ZERO_POINTS | arg_id,
                /*is_input=*/true, /*is_output=*/false, zp_layout);
    };
    if (zp_cfg.do_src_compensation && zp_cfg.is_runtime_src_zero_points) {
        if (zp_cfg.needs_src_conv_precalc) {
            gpu_assert(zp_src);
            int arg_key = DNNL_ARG_ATTR_ZERO_POINTS | DNNL_ARG_SRC;
            tensor_cfg.add_tensor("src_zero_points", arg_key, /*is_input=*/true,
                    /*is_output=*/false, make_layout(*zp_src), layout_t());
        } else {
            add_zp_buffer("src_zero_points", zp_cfg.src_zp_type, DNNL_ARG_SRC,
                    (zp_cfg.is_common_src_zero_point) ? 1 : ic);
        }
    }
    if (zp_cfg.do_wei_compensation && zp_cfg.is_runtime_wei_zero_points) {
        gpu_assert(zp_cfg.is_common_wei_zero_point);
        add_zp_buffer(
                "wei_zero_points", zp_cfg.wei_zp_type, DNNL_ARG_WEIGHTS, 1);
    }
    if (zp_cfg.do_dst_compensation && zp_cfg.is_runtime_dst_zero_points) {
        add_zp_buffer("dst_zero_points", zp_cfg.dst_zp_type, DNNL_ARG_DST, oc);
    }
    auto scale_args = get_scale_args();
    for (int i = 0; i < (int)scale_args.size(); i++) {
        int arg = scale_args[i].second;
        if (attr.scales_.has_default_values(arg)) continue;
        std::vector<dim_t> dims = {(attr.scales_.get_mask(arg) == 0) ? 1 : oc};
        layout_t layout(dsl::type_t::f32(), dims);
        int arg_key = DNNL_ARG_ATTR_SCALES | arg;
        tensor_cfg.add_tensor(scale_args[i].first, arg_key, /*is_input=*/true,
                /*is_output=*/false, layout);
    }
    for (int i = 0; i < attr.post_ops_.len(); i++) {
        auto &po = attr.post_ops_.entry_[i];
        if (po.is_eltwise()
                || po.is_sum(/*require_scale_one=*/false,
                        /*require_zp_zero=*/false)) {
            // No extra tensors.
        } else if (po.is_binary()) {
            auto layout = make_layout(po.binary.src1_desc);
            int arg_key = DNNL_ARG_ATTR_MULTIPLE_POST_OP(i) | DNNL_ARG_SRC_1;
            tensor_cfg.add_tensor("binary_rhs_" + std::to_string(i), arg_key,
                    /*is_input=*/true,
                    /*is_output=*/false, layout);
        } else if (po.is_prelu()) {
            layout_t layout(dsl::type_t::f32(),
                    get_prelu_weights_dims(po.prelu.mask, dst_md));
            int arg_key = DNNL_ARG_ATTR_MULTIPLE_POST_OP(i) | DNNL_ARG_WEIGHTS;
            tensor_cfg.add_tensor("prelu_rhs_" + std::to_string(i), arg_key,
                    /*is_input=*/true, /*is_output=*/false, layout);
        } else {
            gpu_error_not_expected();
        }
    }
}

std::vector<std::pair<char, dim_t>> parse_letter_blocks(
        const std::string &format) {
    std::vector<std::pair<char, dim_t>> ret;

    stringstream_t ss(format);
    while (!ss.eof()) {
        int next = ss.peek();
        if (ss.eof()) break;
        dim_t block = 0;
        while (std::isdigit(next)) {
            block = 10 * block + (next - '0');
            ss.ignore(1);
            next = ss.peek();
        }
        char letter = char(ss.peek());
        gpu_assert(!ss.eof()) << "EOF is unexpected.";
        ss.ignore(1);
        ret.emplace_back(letter, block);
    }
    return ret;
}

std::vector<layout_block_t> parse_format(
        const std::string &format, int ndims_hint) {
    bool seen_letters[DNNL_MAX_NDIMS] = {};
    int letter_ndims = 0;
    for (char c = 'a'; c < 'a' + DNNL_MAX_NDIMS; c++) {
        if (format.find(c) != std::string::npos) {
            seen_letters[c - 'a'] = true;
            MAYBE_UNUSED(seen_letters);
            letter_ndims++;
        }
    }

    for (int i = 0; i < DNNL_MAX_NDIMS; i++) {
        gpu_assert(seen_letters[i] == (i < letter_ndims));
    }

    auto letter_blocks = parse_letter_blocks(format);

    std::vector<layout_block_t> parts;
    for (int i = into<int>(letter_blocks.size() - 1); i >= 0; i--) {
        char letter = letter_blocks[i].first;
        dim_t block = letter_blocks[i].second;
        if (letter != 'x') {
            int dim_idx = std::tolower(letter) - 'a';
            parts.emplace_back(dim_idx, block);
        } else {
            gpu_assert(ndims_hint >= letter_ndims);
            for (int i = ndims_hint - 1; i >= letter_ndims; i--) {
                parts.emplace_back(i, 0);
            }
        }
    }

    return parts;
}

bool matches_tag(
        const layout_t &layout, const std::string &tag, const tile_t &dims) {
    if (layout.is_empty()) return false;
    auto tag_layout = make_layout(layout.type(), dims, tag);
    if (!layout.is_equal_normalized(tag_layout)) return false;
    return true;
}

} // namespace jit
} // namespace intel
} // namespace gpu
} // namespace impl
} // namespace dnnl