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
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186

/*******************************************************************************
* Copyright 2019 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 "common/reorder.hpp"
#include "common/utils.hpp"

#include "gpu/generic/cross_engine_reorder.hpp"
#include "gpu/gpu_engine.hpp"
#include "gpu/gpu_stream.hpp"
#include "gpu/gpu_utils.hpp"

namespace dnnl {
namespace impl {
namespace gpu {
namespace generic {

void cross_engine_reorder_t::pd_t::init_scratchpad(impl::engine_t *gpu_engine) {
    if (do_reorder_) {
        using namespace memory_tracking::names;
        auto gpu_align = utils::downcast<gpu::engine_t *>(gpu_engine)
                                 ->get_buffer_alignment();
        auto scratchpad = scratchpad_registry().registrar();
        auto needs_dst = desc()->src_engine_kind == reorder_engine_kind_;
        memory_desc_wrapper wspace((needs_dst) ? dst_md() : src_md());
        scratchpad.book(key_reorder_cross_space, wspace.size(), 1, gpu_align);
        scratchpad.book(key_nested, reorder_pd_->scratchpad_registry().size(),
                1, gpu_align);
    }
}

status_t cross_engine_reorder_t::pd_t::init(impl::engine_t *engine,
        impl::engine_t *src_engine, impl::engine_t *dst_engine) {
    VDISPATCH_REORDER(src_engine != dst_engine, VERBOSE_BAD_ENGINE_KIND);
    VDISPATCH_REORDER(utils::one_of(engine_kind::gpu, src_engine->kind(),
                              dst_engine->kind()),
            VERBOSE_BAD_ENGINE_KIND);
    VDISPATCH_REORDER(attr_ok(), VERBOSE_UNSUPPORTED_ATTR);
    VDISPATCH_REORDER(extra_ok(true), VERBOSE_UNSUPPORTED_MD_FLAG, "extra_ok");
    VDISPATCH_REORDER(impl::is_dense_format_kind({src_md(), dst_md()}),
            VERBOSE_UNSUPPORTED_SPARSE_CFG);

    memory_desc_wrapper src_mdw(src_md());
    memory_desc_wrapper dst_mdw(dst_md());

    VDISPATCH_REORDER(!src_mdw.has_runtime_dims_or_strides(),
            VERBOSE_RUNTIMEDIM_UNSUPPORTED);

    gpu::quantization_t src_quant {attr(), src_mdw, DNNL_ARG_SRC};
    gpu::quantization_t dst_quant {attr(), dst_mdw, DNNL_ARG_DST};
    gpu::sum_quantization_t sum_quant {attr()};
    bool with_sum_ab = src_quant.with_scale() || src_quant.with_zp()
            || dst_quant.with_scale() || dst_quant.with_zp()
            || sum_quant.with_scale() || sum_quant.with_zp();
    do_reorder_ = with_sum_ab || src_mdw != dst_mdw;

    impl::engine_t *reorder_engine
            = src_engine->kind() == engine_kind::gpu ? src_engine : dst_engine;

    primitive_attr_t r_attr(*attr());
    if (!r_attr.is_initialized()) return status::out_of_memory;

    auto clean_src_md = *src_md();
    auto clean_dst_md = *dst_md();
    clean_src_md.extra = clean_dst_md.extra = {};
    VDISPATCH_REORDER_SC(
            reorder_primitive_desc_create(reorder_pd_, reorder_engine,
                    &clean_src_md, &clean_dst_md, &r_attr),
            VERBOSE_PRIMITIVE_CREATION_FAIL, "reorder");

    reorder_pd_t::init_desc(
            src_engine->kind(), dst_engine->kind(), true /* is_cross_engine */);

    VDISPATCH_REORDER_SC(maybe_create_zp_precompute_conv_pd(dst_engine),
            "failed to create nested zp precompute convolution");
    init_scratchpad(
            (dst_engine->kind() == engine_kind::gpu) ? dst_engine : src_engine);
    return status::success;
}

status_t cross_engine_reorder_t::init(impl::engine_t *engine) {
    CHECK(pd()->maybe_create_zp_precompute_conv(
            zp_precomp_conv_, engine, this));
    if (!pd()->do_reorder_) return status::success;
    return create_nested_primitive(reorder_, pd()->reorder_pd_, engine);
}

status_t cross_engine_reorder_t::execute(const exec_ctx_t &ctx) const {
    using namespace memory_tracking::names;
    auto *gpu_stream = utils::downcast<gpu::stream_t *>(ctx.stream());

    status_t status = status::success;

    auto &src = CTX_IN_STORAGE(DNNL_ARG_FROM);
    auto &dst = CTX_OUT_STORAGE(DNNL_ARG_TO);

    std::unique_ptr<memory_t, memory_deleter_t> wspace;
    if (pd()->do_reorder_) {
        auto src_engine_kind = pd()->desc()->src_engine_kind;
        auto reorder_engine_kind = pd()->reorder_engine_kind_;
        auto scratchpad = ctx.get_scratchpad_grantor().get_memory_storage(
                key_reorder_cross_space);
        auto wspace_md = src_engine_kind == reorder_engine_kind
                ? pd()->dst_md()
                : pd()->src_md();
        CHECK(safe_ptr_assign(wspace,
                new memory_t(ctx.stream()->engine(), wspace_md,
                        std::move(scratchpad))));
    }

    auto exec_reorder = [&](const memory_t *src_mem, const memory_t *dst_mem,
                                const memory_t *src_scales_mem,
                                const memory_t *dst_scales_mem) {
        exec_args_t r_args;
        r_args[DNNL_ARG_SRC]
                = memory_arg_t {const_cast<memory_t *>(src_mem), true};
        r_args[DNNL_ARG_DST]
                = memory_arg_t {const_cast<memory_t *>(dst_mem), false};
        r_args[DNNL_ARG_ATTR_SCALES | DNNL_ARG_SRC]
                = memory_arg_t {const_cast<memory_t *>(src_scales_mem), true};
        r_args[DNNL_ARG_ATTR_SCALES | DNNL_ARG_DST]
                = memory_arg_t {const_cast<memory_t *>(dst_scales_mem), true};

        exec_ctx_t r_ctx(ctx, std::move(r_args));

        auto *nested_grantor
                = create_nested_grantor(ctx.get_scratchpad_grantor(),
                        key_nested, reorder_->pd()->scratchpad_registry());
        r_ctx.set_scratchpad_grantor(nested_grantor);
        return reorder_->execute(r_ctx);
    };

    if (pd()->desc()->src_engine_kind == engine_kind::gpu) {
        // GPU -> CPU or GPU -> GPU
        memory_desc_wrapper dst_mdw(pd()->dst_md());
        if (pd()->do_reorder_) {
            if (pd()->beta() != 0.f) {
                status = gpu_stream->copy(dst, *wspace->memory_storage(),
                        dst_mdw.size(), gpu_stream->ctx().get_deps(),
                        gpu_stream->ctx().get_deps());
            }
            if (status == status::success)
                status = exec_reorder(ctx.input(DNNL_ARG_FROM), wspace.get(),
                        ctx.input(DNNL_ARG_ATTR_SCALES | DNNL_ARG_SRC),
                        ctx.input(DNNL_ARG_ATTR_SCALES | DNNL_ARG_DST));
        }
        if (status == status::success) {
            status = gpu_stream->copy(
                    pd()->do_reorder_ ? *wspace->memory_storage() : src, dst,
                    dst_mdw.size(), gpu_stream->ctx().get_deps(),
                    gpu_stream->ctx().get_deps());
        }
    } else {
        // CPU -> GPU
        memory_desc_wrapper src_mdw(pd()->src_md());
        status = gpu_stream->copy(src,
                pd()->do_reorder_ ? *wspace->memory_storage() : dst,
                src_mdw.size(), gpu_stream->ctx().get_deps(),
                gpu_stream->ctx().get_deps());
        if (status == status::success && pd()->do_reorder_) {
            status = exec_reorder(wspace.get(), ctx.output(DNNL_ARG_TO),
                    ctx.input(DNNL_ARG_ATTR_SCALES | DNNL_ARG_SRC),
                    ctx.input(DNNL_ARG_ATTR_SCALES | DNNL_ARG_DST));
        }
        if (status == status::success)
            status = pd()->maybe_exec_zp_precompute_conv(ctx, zp_precomp_conv_);
    }
    return status;
}

} // namespace generic
} // namespace gpu
} // namespace impl
} // namespace dnnl