inference-engine/thirdparty/mkl-dnn/src/cpu/gemm_u8s8s32x_inner_product.hpp

   1 /*******************************************************************************
   2 * Copyright 2018 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 GEMM_U8S8S32X_INNER_PRODUCT_HPP
  18 #define GEMM_U8S8S32X_INNER_PRODUCT_HPP
  19
  20 #include <assert.h>
  21
  22 #include "c_types_map.hpp"
  23 #include "cpu_inner_product_pd.hpp"
  24 #include "cpu_engine.hpp"
  25 #include "type_helpers.hpp"
  26 #include "utils.hpp"
  27 #include "scratchpad.hpp"
  28
  29 #include "gemm/os_blas.hpp"
  30
  31 namespace mkldnn {
  32 namespace impl {
  33 namespace cpu {
  34
  35 template <impl::data_type_t dst_type>
  36 struct gemm_u8s8s32x_inner_product_fwd_t: public cpu_primitive_t {
  37     struct pd_t: public cpu_inner_product_fwd_pd_t {
  38         pd_t(engine_t *engine, const inner_product_desc_t *adesc,
  39                 const primitive_attr_t *attr,
  40                 const inner_product_fwd_pd_t *hint_fwd_pd)
  41             : cpu_inner_product_fwd_pd_t(engine, adesc, attr, hint_fwd_pd) {}
  42
  43         DECLARE_COMMON_PD_T("gemm:blas", gemm_u8s8s32x_inner_product_fwd_t);
  44
  45         virtual status_t init() override {
  46             using namespace utils;
  47             using namespace data_type;
  48
  49             assert(engine()->kind() == engine_kind::cpu);
  50
  51             bool ok = true
  52 #if !USE_MKL_IGEMM
  53                 && false
  54 #endif
  55                 && this->set_default_params() == status::success
  56                 && one_of(desc()->prop_kind, prop_kind::forward_training,
  57                         prop_kind::forward_inference)
  58                 && !has_zero_dim_memory()
  59                 && this->desc()->src_desc.data_type == u8
  60                 && this->desc()->dst_desc.data_type == dst_type
  61                 && this->desc()->weights_desc.data_type == s8
  62                 && IMPLICATION(this->with_bias(), utils::one_of(
  63                             this->desc()->bias_desc.data_type, f32, s32, s8,
  64                             u8))
  65                 && attr()->post_ops_.len_ <= 1
  66                 && IMPLICATION(attr()->post_ops_.len_,
  67                         attr()->post_ops_.entry_[0].is_relu(true, false))
  68                 && dense_gemm_consitency_check(src_pd(), weights_pd(),
  69                         dst_pd());
  70             return ok ? status::success : status::unimplemented;
  71         }
  72
  73     protected:
  74         virtual status_t set_default_params() override {
  75             using namespace memory_format;
  76
  77             if (this->src_pd_.desc()->format == any)
  78             {
  79                 if (ndims() == 4) CHECK(this->src_pd_.set_format(nhwc));
  80                 else if (ndims() == 5) CHECK(this->src_pd_.set_format(ndhwc));
  81                 else CHECK(this->src_pd_.set_format(nc));
  82             }
  83             if (this->dst_pd_.desc()->format == any)
  84                 CHECK(this->dst_pd_.set_format(nc));
  85             if (this->weights_pd_.desc()->format == any)
  86             {
  87                 if (ndims() == 4) CHECK(this->weights_pd_.set_format(hwio));
  88                 else if (ndims() == 5) CHECK(this->weights_pd_.set_format(dhwio));
  89                 else CHECK(this->weights_pd_.set_format(io));
  90             }
  91             if (this->bias_pd_.desc()->format == any)
  92                 CHECK(this->bias_pd_.set_format(x));
  93             return status::success;
  94         }
  95     };
  96
  97     gemm_u8s8s32x_inner_product_fwd_t(const pd_t *pd, const input_vector &inputs,
  98             const output_vector &outputs)
  99         : cpu_primitive_t(&conf_, inputs, outputs), conf_(*pd), dst_is_acc_(false),
 100         scratchpad_(nullptr)
 101     {
 102         dst_is_acc_ = utils::one_of(dst_type, data_type::s32, data_type::f32);
 103         if (!dst_is_acc_) {
 104             size_t size = conf_.MB() * conf_.OC() * sizeof(acc_data_t);
 105             scratchpad_ = create_scratchpad(size);
 106         }
 107     }
 108     ~gemm_u8s8s32x_inner_product_fwd_t() { delete scratchpad_; };
 109
 110     typedef typename prec_traits<dst_type>::type data_t;
 111
 112     typedef typename prec_traits<data_type::u8>::type src_data_t;
 113     typedef typename prec_traits<data_type::s8>::type wei_data_t;
 114     typedef typename prec_traits<dst_type>::type dst_data_t;
 115     typedef typename prec_traits<data_type::s32>::type acc_data_t;
 116
 117     virtual void execute(event_t *e) {
 118         execute_forward();
 119         e->set_state(event_t::ready);
 120     }
 121
 122 private:
 123     void execute_forward();
 124     pd_t conf_;
 125     bool dst_is_acc_;
 126     scratchpad_t *scratchpad_;
 127 };
 128 }
 129 }
 130 }
 131
 132 #endif
 133
 134 // vim: et ts=4 sw=4 cindent cino^=l0,\:0,N-s