pre_processing_xe.hpp

Go to the documentation of this file.

/*******************************************************************************
* Copyright (c) 2022-2023 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.
*******************************************************************************/
 
 
#pragma once
 
#include "group/gemm/api.hpp"
#include "group/gemm/common.hpp"
 
namespace gpu::xetla::group {
 
 
template <typename tile_shape_, gpu_arch arch_tag>
class pre_processing_default_t<tile_shape_, arch_tag,
        std::enable_if_t<(arch_tag == gpu_arch::Xe)>> {
    using tile_shape = tile_shape_;
    using work_group_t = typename tile_shape::work_group_t;
 
public:
    struct arguments_t {};
 
    inline pre_processing_default_t() = default;
 
    inline pre_processing_default_t([[maybe_unused]] work_group_t &g,
            [[maybe_unused]] arguments_t &args) {}
 
    inline void init([[maybe_unused]] work_group_t &g,
            [[maybe_unused]] arguments_t &args) {}
 
    template <typename matA_acc_t, typename matB_acc_t, typename matA_t,
            typename matB_t>
    inline KERNEL_FUNC void operator()([[maybe_unused]] matA_acc_t &matA_acc,
            [[maybe_unused]] matB_acc_t &matB_acc,
            [[maybe_unused]] matA_t &matA, [[maybe_unused]] matB_t &matB) {}
};
 
template <typename tile_shape_, gpu_arch arch_tag>
class pre_processing_matA_neg_filter_t<tile_shape_, arch_tag,
        std::enable_if_t<(arch_tag == gpu_arch::Xe)>> {
    using tile_shape = tile_shape_;
    using work_group_t = typename tile_shape::work_group_t;
 
public:
    struct arguments_t {};
 
    inline pre_processing_matA_neg_filter_t() = default;
 
    inline pre_processing_matA_neg_filter_t([[maybe_unused]] work_group_t &g,
            [[maybe_unused]] arguments_t &args) {}
 
    inline void init([[maybe_unused]] work_group_t &g,
            [[maybe_unused]] arguments_t &args) {}
 
    template <typename matA_acc_t, typename matB_acc_t, typename matA_t,
            typename matB_t>
    inline KERNEL_FUNC void operator()([[maybe_unused]] matA_acc_t &matA_acc,
            [[maybe_unused]] matB_acc_t &matB_acc,
            [[maybe_unused]] matA_t &matA, [[maybe_unused]] matB_t &matB) {
 
        using data_t = typename matA_acc_t::dtype;
        if constexpr (sizeof(data_t) == 2) {
            xetla_mask<matA_acc_t::tile_elems> mask
                    = matA_acc.reg.xetla_format<int16_t>() < 0;
            matA_acc.reg.xetla_format<int16_t>().xetla_merge(0, mask);
        }
        if constexpr (sizeof(data_t) == 1) {
            xetla_mask<matA_acc_t::tile_elems> mask
                    = matA_acc.reg.xetla_format<int8_t>() < 0;
            matA_acc.reg.xetla_format<int8_t>().xetla_merge(0, mask);
        }
        if constexpr (sizeof(data_t) == 4) {
            xetla_mask<matA_acc_t::tile_elems> mask
                    = matA_acc.reg.xetla_format<int32_t>() < 0;
            matA_acc.reg.xetla_format<int32_t>().xetla_merge(0, mask);
        }
    }
};
 
 
} // namespace gpu::xetla::group