ref_deconvolution.hpp 21 KB
Newer Older
Alexey Suhov's avatar
Alexey Suhov committed
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521
/*******************************************************************************
* Copyright 2018-2019 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.
*******************************************************************************/

#ifndef CPU_REF_DECONVOLUTION_HPP
#define CPU_REF_DECONVOLUTION_HPP

#include <assert.h>
#include <string.h>

#include "c_types_map.hpp"
#include "cpu_convolution_pd.hpp"
#include "cpu_deconvolution_pd.hpp"
#include "cpu_engine.hpp"
#include "type_helpers.hpp"
#include "utils.hpp"
#include "primitive_iterator.hpp"

namespace mkldnn {
namespace impl {
namespace cpu {

static status_t compute_blocked_format(bool with_groups,
    const memory_desc_t *oi_md, memory_desc_t *io_md)
{
    using namespace memory_format;
    using namespace types;
    /* Computes blocking for *i*o* format from *o*i* format */
    if (oi_md->ndims != io_md->ndims) return status::invalid_arguments;
    blocking_desc_t oi_blk = oi_md->layout_desc.blocking,
        &io_blk = io_md->layout_desc.blocking;
    io_blk = oi_blk;
    nstl::swap(io_blk.strides[0][0+with_groups], io_blk.strides[0][1+with_groups]);
    nstl::swap(io_blk.strides[1][0+with_groups], io_blk.strides[1][1+with_groups]);
    nstl::swap(io_blk.padding_dims[0+with_groups], io_blk.padding_dims[1+with_groups]);
    nstl::swap(io_blk.offset_padding_to_data[0+with_groups],
         io_blk.offset_padding_to_data[1+with_groups]);
    nstl::swap(io_blk.block_dims[0+with_groups], io_blk.block_dims[1+with_groups]);

    if (is_format_double_blocked(oi_md->format)) {
        memory_format_t fmt;
        switch (oi_md->format) {
        case gOIhw8o16i2o: fmt = gIOhw8i16o2i; break;
        case OIhw8o16i2o: fmt = IOhw8i16o2i; break;
        /* 1x1 formats */
        case IOhw8o16i2o: fmt = OIhw8i16o2i; break;
        case gIOhw8o16i2o: fmt = gOIhw8i16o2i; break;
        case gOIhw8i16o2i: fmt = gIOhw8o16i2o; break;
        case OIhw8i16o2i: fmt = IOhw8o16i2o; break;
        default: return unimplemented;
        }
        io_md->format = fmt;
    } else
        io_md->format = memory_format::blocked;
    return status::success;
}

static status_t conv_descr_create(const deconvolution_desc_t *dd,
        convolution_desc_t *cd)
{
    using namespace prop_kind;
    using namespace memory_format;
    alg_kind_t alg_kind = ( dd->alg_kind == alg_kind::deconvolution_direct
        ? alg_kind::convolution_direct : alg_kind::convolution_winograd );
    prop_kind_t prop_kind;
    const memory_desc_t *src_md, *dst_md;
    memory_desc_t c_weights_d, d_weights_d;
    bool with_groups;
    if ( utils::one_of(dd->prop_kind, forward_training, forward_inference) ) {
        prop_kind = backward_data;
        src_md = &dd->dst_desc;
        dst_md = &dd->src_desc;
        d_weights_d = dd->weights_desc;
    } else if (dd->prop_kind == backward_data) {
        prop_kind = forward_training;
        src_md = &dd->diff_dst_desc;
        dst_md = &dd->diff_src_desc;
        d_weights_d = dd->weights_desc;
    } else {
        prop_kind = dd->prop_kind;
        src_md = &dd->diff_dst_desc;
        dst_md = &dd->src_desc;
        d_weights_d = dd->diff_weights_desc;
    }
    with_groups = d_weights_d.ndims == src_md->ndims + 1;

    /* create weights desc for convolution */
    c_weights_d = d_weights_d;
    nstl::swap(c_weights_d.dims[with_groups + 0], c_weights_d.dims[with_groups + 1]);
    if (c_weights_d.format != any)
    {
        if (utils::one_of(c_weights_d.format, gOIhw4i16o4i, OIhw4i16o4i))
            return unimplemented;
        CHECK(compute_blocked_format(with_groups, &d_weights_d, &c_weights_d));
    }
    return conv_desc_init(cd, prop_kind, alg_kind, src_md, &(c_weights_d),
            (prop_kind != backward_weights ? &(dd->bias_desc) : nullptr),
            dst_md, dd->strides, dd->dilates,
            dd->padding[0], dd->padding[1], dd->padding_kind);
}

struct ref_deconvolution_fwd_t: public cpu_primitive_t {
    struct pd_t: public cpu_deconvolution_fwd_pd_t {
        pd_t(engine_t *engine,
                const deconvolution_desc_t *adesc,
                const primitive_attr_t *attr,
                const deconvolution_fwd_pd_t *hint_fwd_pd)
            : cpu_deconvolution_fwd_pd_t(engine, adesc, attr, hint_fwd_pd)
            , conv_pd_(nullptr)
        {}

        pd_t(const pd_t &other)
            : cpu_deconvolution_fwd_pd_t(other)
            , conv_pd_(other.conv_pd_->clone())
            , conv_supports_bias_(other.conv_supports_bias_)
        {}

        ~pd_t() { delete conv_pd_; }

        DECLARE_DECONVOLUTION_PD_T(ref_deconvolution_fwd_t);

        status_t init_convolution(){
            using namespace memory_format;
            using namespace types;
            convolution_desc_t cd;
            status_t status;

            status = conv_descr_create(this->desc(), &cd);
            if (status != status::success) return status;

            mkldnn_primitive_desc_iterator it(this->engine_, (op_desc_t *)&cd,
                &(this->attr_), nullptr);
            while (++it != it.end()) {
                conv_pd_ = *it;
                conv_supports_bias_ = static_cast<cpu_convolution_bwd_data_pd_t *>
                    (conv_pd_)->support_bias();
                bool bias_supported = true
                        && desc()->accum_data_type == data_type::f32
                        && utils::one_of(desc()->dst_desc.data_type,
                                           data_type::f32, data_type::bf16);
                auto wei_fmt = format_normalize(conv_pd_->weights_pd()->desc()->format);
                auto src_fmt = conv_pd_->diff_dst_pd()->desc()->format;

                bool ok = true
                        && (wei_fmt == blocked)
                        && IMPLICATION(
                                desc()->src_desc.data_type == data_type::bf16,
                                utils::one_of(src_fmt,
                                    nCw16c, nChw16c, nCdhw16c))
                        && IMPLICATION(with_bias(),
                                   conv_supports_bias_ || bias_supported);
                if (ok)
                    return success;
                delete conv_pd_;
            }
            conv_pd_ = nullptr;
            return unimplemented;
        };
        virtual status_t init() override {
            using namespace prop_kind;
            assert(this->engine()->kind() == engine_kind::cpu);
            bool ok = true
                && utils::one_of(this->desc()->prop_kind, forward_training,
                        forward_inference)
                && utils::one_of(this->desc()->alg_kind,
                        alg_kind::deconvolution_direct,
                        alg_kind::deconvolution_winograd)
                && attr()->post_ops_.has_default_values();

            if (ok) {
                CHECK(init_convolution());
                if (weights_pd_.desc()->format == memory_format::any) {
                    CHECK(compute_blocked_format(with_groups(),
                            conv_pd_->weights_pd()->desc(),
                            &desc_.weights_desc));
                    cpu_memory_pd_t weights(engine_, &desc_.weights_desc);
                    weights_pd_ = weights;
                }
                if (src_pd_.desc()->format == memory_format::any)
                    CHECK(src_pd_.set_format(conv_pd_->diff_dst_pd()->desc()->format));
                if (dst_pd_.desc()->format == memory_format::any)
                    CHECK(dst_pd_.set_format(conv_pd_->diff_src_pd()->desc()->format));
                if (bias_pd_.desc()->format == memory_format::any)
                    CHECK(bias_pd_.set_format(memory_format::x));
                return status::success;
            }
            else return status::unimplemented;
        }
        primitive_desc_t *conv_pd_;
        bool conv_supports_bias_;
    };

    ref_deconvolution_fwd_t(const pd_t *apd, const input_vector &inputs,
            const output_vector &outputs)
        : cpu_primitive_t(apd, inputs, outputs), conv_p_(nullptr) {}

    ~ref_deconvolution_fwd_t() { delete this->conv_p_; }

    virtual void execute(event_t *e) const {
        switch (pd()->desc()->prop_kind) {
        case prop_kind::forward_training:
        case prop_kind::forward_inference:
            (conv_p_)->execute(e);
            if (pd()->with_bias() && !pd()->conv_supports_bias_) {
                auto dst_t = pd()->desc()->dst_desc.data_type;

                if (dst_t == data_type::bf16) {
                    compute_fwd_bias_nCdhwXc_bf16<16>();
                } else {
                    switch (pd()->dst_pd()->desc()->format) {
                    /* XXX: current implementation only provides funcitonality. This
                     * needs to be cleaned and optimized. */
                    case memory_format::ncw:
                    case memory_format::nchw:
                    case memory_format::ncdhw:
                        compute_fwd_bias_ncdhw();
                        break;
                    case memory_format::nChw8c:
                    case memory_format::nCdhw8c:
                        compute_fwd_bias_nCdhwXc<8>();
                        break;
                    case memory_format::nCw16c:
                    case memory_format::nChw16c:
                    case memory_format::nCdhw16c:
                        compute_fwd_bias_nCdhwXc<16>();
                        break;
                    default:
                        compute_fwd_bias();
                        break;
                    }
                }
            }
            break;
        default:
            assert(!"invalid prop_kind");
        }
        e->set_state(event_t::ready);
    }

private:
    typedef typename prec_traits<data_type::f32>::type f32_data_t;
    typedef typename prec_traits<data_type::bf16>::type bf16_data_t;
    void compute_fwd_bias() const;
    void compute_fwd_bias_ncdhw() const;
    template <int blksize> void compute_fwd_bias_nCdhwXc() const;
    template <int blksize> void compute_fwd_bias_nCdhwXc_bf16() const;
    const pd_t *pd() const { return (const pd_t *)primitive_t::pd(); }
    primitive_t *conv_p_;
};

struct ref_deconvolution_bwd_data_t: public cpu_primitive_t {
    struct pd_t: public cpu_deconvolution_bwd_data_pd_t {
        pd_t(engine_t *engine,
                const deconvolution_desc_t *adesc,
                const primitive_attr_t *attr,
                const deconvolution_fwd_pd_t *hint_fwd_pd)
            : cpu_deconvolution_bwd_data_pd_t(engine, adesc, attr, hint_fwd_pd)
            , conv_pd_(nullptr)
        {}

        pd_t(const pd_t &other)
            : cpu_deconvolution_bwd_data_pd_t(other)
            , conv_pd_(other.conv_pd_->clone()) {}

        ~pd_t() { delete conv_pd_; }

        DECLARE_DECONVOLUTION_PD_T(ref_deconvolution_bwd_data_t);

        status_t init_convolution(){
            using namespace memory_format;
            using namespace types;
            convolution_desc_t cd;
            status_t status;

            status = conv_descr_create(this->desc(), &cd);
            if (status != status::success) return status;

             mkldnn_primitive_desc_iterator it(this->engine_, (op_desc_t *)&cd,
                &(this->attr_), nullptr);
             while (++it != it.end()) {
                conv_pd_ = *it;
                if (format_normalize(conv_pd_->weights_pd()->desc()->format)
                        == blocked)
                    return success;
                delete conv_pd_;
            }
            conv_pd_ = nullptr;
            return unimplemented;
        };

        virtual status_t init() override {
            using namespace prop_kind;
            using namespace data_type;
            assert(this->engine()->kind() == engine_kind::cpu);
            bool ok = true
                && this->desc()->prop_kind == backward_data
                && (utils::everyone_is(data_type::f32,
                                this->desc()->weights_desc.data_type,
                                this->desc()->diff_dst_desc.data_type)
                        || utils::everyone_is(data_type::bf16,
                            this->desc()->weights_desc.data_type,
                            this->desc()->diff_dst_desc.data_type))
                && utils::one_of(
                        this->desc()->diff_src_desc.data_type,
                        data_type::f32,
                        data_type::bf16)
                && utils::one_of(this->desc()->alg_kind,
                               alg_kind::deconvolution_direct,
                               alg_kind::deconvolution_winograd);

            if (ok) {
                CHECK(init_convolution());
                if (weights_pd_.desc()->format == memory_format::any)
                {
                    CHECK(compute_blocked_format(with_groups(),
                        conv_pd_->weights_pd()->desc(),
                        &desc_.weights_desc));
                    cpu_memory_pd_t weights(engine_, &desc_.weights_desc);
                    weights_pd_ = weights;
                }
                if (diff_src_pd_.desc()->format == memory_format::any)
                    CHECK(diff_src_pd_.set_format(conv_pd_->dst_pd()->desc()->format));
                if (diff_dst_pd_.desc()->format == memory_format::any)
                    CHECK(diff_dst_pd_.set_format(conv_pd_->src_pd()->desc()->format));
                return status::success;
            }
            else return status::unimplemented;
        }
        primitive_desc_t *conv_pd_;
    };
    ref_deconvolution_bwd_data_t(const pd_t *apd, const input_vector &inputs,
            const output_vector &outputs)
        : cpu_primitive_t(apd, inputs, outputs), conv_p_(nullptr) {}
    ~ref_deconvolution_bwd_data_t() { delete this->conv_p_; }

    virtual void execute(event_t *e) const {
        switch (pd()->desc()->prop_kind) {
        case prop_kind::backward_data:
            (conv_p_)->execute(e);
            break;
        default:
            assert(!"invalid prop_kind");
        }
        e->set_state(event_t::ready);
    }

private:
    const pd_t *pd() const { return (const pd_t *)primitive_t::pd(); }
    primitive_t *conv_p_;
};

struct ref_deconvolution_bwd_weights_t: public cpu_primitive_t {
    struct pd_t: public cpu_deconvolution_bwd_weights_pd_t {
        pd_t(engine_t *engine,
                const deconvolution_desc_t *adesc,
                const primitive_attr_t *attr,
                const deconvolution_fwd_pd_t *hint_fwd_pd)
            : cpu_deconvolution_bwd_weights_pd_t(engine, adesc, attr, hint_fwd_pd)
            , conv_pd_(nullptr)
        {}

        pd_t(const pd_t &other)
            : cpu_deconvolution_bwd_weights_pd_t(other)
            , conv_pd_(other.conv_pd_->clone()) {}

        ~pd_t() { delete conv_pd_; }

        DECLARE_DECONVOLUTION_PD_T(ref_deconvolution_bwd_weights_t);

        status_t init_convolution(){
            using namespace memory_format;
            using namespace types;
            convolution_desc_t cd;
            status_t status;

            status = conv_descr_create(this->desc(), &cd);
            if (status != status::success) return status;

             mkldnn_primitive_desc_iterator it(this->engine_, (op_desc_t *)&cd,
                &(this->attr_), nullptr);
             while (++it != it.end()) {
                conv_pd_ = *it;
                auto wei_fmt = conv_pd_->diff_weights_pd()->desc()->format;
                auto diff_dst_fmt = conv_pd_->src_pd()->desc()->format;
                bool ok = true
                        && format_normalize(wei_fmt) == blocked
                        && !is_format_double_blocked(wei_fmt)
                        && IMPLICATION(
                                desc()->src_desc.data_type == data_type::bf16,
                                utils::one_of(diff_dst_fmt,
                                    nCw16c, nChw16c, nCdhw16c));
                if (ok)
                    return success;
                delete conv_pd_;
            }
            conv_pd_ = nullptr;
            return unimplemented;
        };

        virtual status_t init() override {
            using namespace prop_kind;
            assert(this->engine()->kind() == engine_kind::cpu);
            bool ok = true
                && this->desc()->prop_kind == backward_weights
                && (utils::everyone_is(data_type::f32,
                        this->desc()->src_desc.data_type,
                        this->desc()->diff_dst_desc.data_type)
                    || (utils::everyone_is(data_type::bf16,
                            this->desc()->src_desc.data_type,
                            this->desc()->diff_dst_desc.data_type)
                        && utils::one_of(
                            this->desc()->diff_weights_desc.data_type,
                                                     data_type::f32,
                                                     data_type::bf16)))
                && utils::one_of(this->desc()->alg_kind,
                        alg_kind::deconvolution_direct,
                        alg_kind::deconvolution_winograd)
                && this->attr()->has_default_values()
                && IMPLICATION(this->with_bias(),
                        (this->desc()->diff_bias_desc.data_type
                                == data_type::f32)
                        && utils::one_of(this->desc()->diff_dst_desc.data_type,
                            data_type::f32, data_type::bf16));
            if (ok) {
                CHECK(init_convolution());
                if (diff_weights_pd_.desc()->format == memory_format::any)
                {
                    CHECK(compute_blocked_format(with_groups(),
                        conv_pd_->diff_weights_pd()->desc(),
                        &desc_.diff_weights_desc));
                    cpu_memory_pd_t weights(engine_, &desc_.diff_weights_desc);
                    diff_weights_pd_ = weights;
                }
                if (src_pd_.desc()->format == memory_format::any)
                    CHECK(src_pd_.set_format(conv_pd_->diff_dst_pd()->desc()->format));
                if (diff_dst_pd_.desc()->format == memory_format::any)
                    CHECK(diff_dst_pd_.set_format(conv_pd_->src_pd()->desc()->format));
                if (diff_bias_pd_.desc()->format == memory_format::any)
                    CHECK(diff_bias_pd_.set_format(memory_format::x));
                return status::success;
            }
            else return status::unimplemented;
        }
        primitive_desc_t *conv_pd_;
    };

    ref_deconvolution_bwd_weights_t(const pd_t *apd, const input_vector &inputs,
            const output_vector &outputs)
        : cpu_primitive_t(apd, inputs, outputs), conv_p_(nullptr) {}

    ~ref_deconvolution_bwd_weights_t() { delete this->conv_p_; }

    typedef typename prec_traits<data_type::f32>::type data_t;

    virtual void execute(event_t *e) const {
        switch (pd()->desc()->prop_kind) {
        case prop_kind::backward_weights:
            (conv_p_)->execute(e);
            if (pd()->with_bias()) {
                auto dst_t = pd()->desc()->diff_dst_desc.data_type;
                if (dst_t == data_type::bf16) {
                    compute_bwd_bias_nCdhwXc_bf16<16>();
                } else {
                    switch (pd()->diff_dst_pd()->desc()->format) {
                    /* XXX: current implementation only provides funcitonality. This
                     * needs to be cleaned and optimized. */
                    case memory_format::ncw:
                    case memory_format::nchw:
                    case memory_format::ncdhw:
                        compute_bwd_bias_ncdhw();
                        break;
                    case memory_format::nChw8c:
                        compute_bwd_bias_nCdhwXc<8>();
                        break;
                    case memory_format::nCw16c:
                    case memory_format::nChw16c:
                    case memory_format::nCdhw16c:
                        compute_bwd_bias_nCdhwXc<16>();
                        break;
                    default:
                        compute_bwd_bias();
                        break;
                    }
                }
            }
            break;
        default:
            assert(!"invalid prop_kind");
        }
        e->set_state(event_t::ready);
    }

private:
    typedef typename prec_traits<data_type::f32>::type f32_data_t;
    typedef typename prec_traits<data_type::bf16>::type bf16_data_t;
    const pd_t *pd() const { return (const pd_t *)primitive_t::pd(); }
    primitive_t *conv_p_;
    void compute_bwd_bias() const;
    void compute_bwd_bias_ncdhw() const;
    template <int blksize> void compute_bwd_bias_nCdhwXc() const;
    template <int blksize> void compute_bwd_bias_nCdhwXc_bf16() const;
};

}
}
}
#endif

// vim: et ts=4 sw=4 cindent cino^=l0,\:0,N-s