TheRock/rocm-libraries/projects/miopen/test/gemm.hpp

/*******************************************************************************
 *
 * MIT License
 *
 * Copyright (c) 2017 Advanced Micro Devices, Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this software and associated documentation files (the "Software"), to deal
 * in the Software without restriction, including without limitation the rights
 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 * copies of the Software, and to permit persons to whom the Software is
 * furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in all
 * copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 *
 *******************************************************************************/
#ifndef GUARD_GEMM_HPP
#define GUARD_GEMM_HPP

#include <iostream>
#include <miopen/ford.hpp>
#include <miopen/errors.hpp>

/*
    A and B rows and cols should be passed as default values (NxM, MxK), independently of
   a_transponse/b_transpose flag value
    C rows and cols should have correct values based on a_transponse/b_transpose values
    A, B, C strides should have corret values based on a_transponse/b_transpose values
*/
template <typename Dtype>
void gemm_cpu(const Dtype* a_ptr,
              const size_t a_cols,
              const size_t a_rows,
              const size_t a_stride,
              const bool a_transpose,
              const Dtype* b_ptr,
              const size_t b_cols,
              const size_t b_rows,
              const size_t b_stride,
              const bool b_transpose,
              Dtype* c_ptr,
              const size_t c_cols,
              const size_t c_rows,
              const size_t c_stride,
              double alpha = 1.0,
              double beta  = 1.0)
{
    if((!a_transpose && !b_transpose &&
        ((a_cols != b_rows) || (a_rows != c_rows) || (b_cols != c_cols))) ||
       (a_transpose && b_transpose &&
        ((a_rows != b_cols) || (a_cols != c_rows) || (b_rows != c_cols))) ||
       (a_transpose && !b_transpose &&
        ((a_rows != b_rows) || (a_cols != c_rows) || (b_cols != c_cols))) ||
       (!a_transpose && b_transpose &&
        ((a_cols != b_cols) || (a_rows != c_rows) || (b_rows != c_cols))))
    {
        MIOPEN_THROW("MM_CPU_ERROR. Incompatible matrix size:\nA: " + std::to_string(a_rows) + "x" +
                     std::to_string(a_cols) + " transpose: " + (a_transpose ? "true" : "false") +
                     "\nB: " + std::to_string(b_rows) + "x" + std::to_string(b_cols) +
                     " transpose: " + (b_transpose ? "true" : "false") +
                     "\nC: " + std::to_string(c_rows) + "x" + std::to_string(c_cols) + "\n");
    }

    size_t inner_loop_limit = a_transpose ? a_rows : a_cols;
    auto inner_loop         = [&](int m, int n) {
        double el = 0.0;
        if(!a_transpose && !b_transpose)
        {
            miopen::ford(inner_loop_limit)([&](int k) {
                el += static_cast<double>(a_ptr[m * a_stride + k]) *
                      static_cast<double>(b_ptr[k * b_stride + n]);
            });
        }
        else if(!a_transpose && b_transpose)
        {
            miopen::ford(inner_loop_limit)([&](int k) {
                el += static_cast<double>(a_ptr[m * a_stride + k]) *
                      static_cast<double>(b_ptr[n * b_stride + k]);
            });
        }
        else if(a_transpose && !b_transpose)
        {
            miopen::ford(inner_loop_limit)([&](int k) {
                el += static_cast<double>(a_ptr[k * a_stride + m]) *
                      static_cast<double>(b_ptr[k * b_stride + n]);
            });
        }
        else
        {
            miopen::ford(inner_loop_limit)([&](int k) {
                el += static_cast<double>(a_ptr[k * a_stride + m]) *
                      static_cast<double>(b_ptr[n * b_stride + k]);
            });
        }

        c_ptr[m * c_stride + n] =
            static_cast<Dtype>(beta * static_cast<double>(c_ptr[m * c_stride + n]) + alpha * el);
    };

    constexpr size_t iter_margin = 1'048'576; // 2^20
    if(c_rows * c_cols * inner_loop_limit > iter_margin)
    {
        miopen::par_ford(c_rows, c_cols)(inner_loop);
    }
    else
    {
        miopen::ford(c_rows, c_cols)(inner_loop);
    }
}

#endif