onednn-src 0.1.13

/*******************************************************************************
* Copyright 2017 Intel Corporation
* Copyright 2018 YANDEX LLC
* Copyright 2020-2024 FUJITSU LIMITED
* Copyright 2025 Arm Ltd. and affiliates
*
* 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 <bitset>

#include "common/dnnl_thread.hpp"

#include "cpu/cpu_pooling_pd.hpp"

#include "cpu/aarch64/jit_uni_pool_kernel.hpp"

namespace dnnl {
namespace impl {
namespace cpu {
namespace aarch64 {

using namespace Xbyak_aarch64;
using namespace alg_kind;

#define GET_OFF(field) \
    static_cast<uint32_t>(offsetof(jit_uni_pooling_args_t, field))

static bcast_set_t get_supported_bcast_strategies() {
    return {broadcasting_strategy_t::scalar, broadcasting_strategy_t::per_oc,
            broadcasting_strategy_t::no_broadcast};
}

template <cpu_isa_t isa>
jit_uni_pool_kernel_t<isa>::~jit_uni_pool_kernel_t() = default;

template <cpu_isa_t isa>
jit_uni_pool_kernel_t<isa>::jit_uni_pool_kernel_t(
        const jit_pool_conf_t &ajpp, const memory_desc_t *dst_md)
    : jit_generator_t(nullptr, MAX_CODE_SIZE, true), jpp(ajpp) {
    if (jpp.with_postops) {
        static constexpr bool preserve_gpr = true;
        static constexpr bool preserve_vmm = true;
        static constexpr bool use_exact_tail_scalar_bcast = false;
        size_t postop_tail = static_cast<size_t>(jpp.c_tail);

        const binary_injector::rhs_arg_static_params_t rhs_sp {
                static_cast<std::size_t>(this->v4.getIdx()), this->x14,
                this->x15, this->x13, preserve_gpr, preserve_vmm,
                GET_OFF(post_ops_binary_rhs_arg_vec), GET_OFF(dst_orig),
                memory_desc_wrapper(jpp.tag_kind == jit_memory_tag_kind_t::ncsp
                                ? jpp.tmp_md
                                : *dst_md),
                postop_tail, k_c_tail_mask_s, use_exact_tail_scalar_bcast};

        const binary_injector::static_params_t bsp {
                reg_param, get_supported_bcast_strategies(), rhs_sp};

        postops_injector_
                = utils::make_unique<injector::jit_uni_postops_injector_t<isa>>(
                        this, jpp.post_ops, bsp);
    }
}

static status_t set_binary_postops_formats(
        post_ops_t &post_ops, const memory_desc_t *dst_md) {
    for (int idx = 0; idx < post_ops.len(); ++idx) {
        if (!post_ops.contain(primitive_kind::binary, idx)) continue;

        auto &src1_md = post_ops.entry_[idx].binary.src1_desc;
        const memory_desc_wrapper src1_mdw(src1_md);
        if (!src1_mdw.format_any()) {
            if (src1_mdw.is_blocking_desc())
                continue;
            else
                return status::unimplemented;
        }

        const memory_desc_wrapper dst_mdw(dst_md);
        assert(!dst_mdw.format_any());

        CHECK(memory_desc_init_by_blocking_desc(
                src1_md, dst_mdw.blocking_desc()));
    }

    return status::success;
}

template <cpu_isa_t isa>
status_t jit_uni_pool_kernel_t<isa>::init_conf(jit_pool_conf_t &jpp,
        memory_tracking::registrar_t &scratchpad, primitive_attr_t &attr,
        const pooling_pd_t *ppd) {

    const auto &pd = *ppd->desc();
    const memory_desc_wrapper src_d(
            ppd->is_fwd() ? ppd->src_md() : ppd->diff_src_md());
    const memory_desc_wrapper dst_d(
            ppd->is_fwd() ? ppd->dst_md() : ppd->diff_dst_md());

    const int ndims = src_d.ndims();

    jpp.nthr = dnnl_get_max_threads();
    jpp.is_training = pd.prop_kind == prop_kind::forward_training;
    jpp.is_backward = pd.prop_kind == prop_kind::backward_data;

    jpp.id = (ndims == 5) ? src_d.dims()[2] : 1;
    jpp.ih = (ndims == 3) ? 1 : src_d.dims()[ndims - 2];
    jpp.iw = src_d.dims()[ndims - 1];
    jpp.od = (ndims == 5) ? dst_d.dims()[2] : 1;
    jpp.ow = dst_d.dims()[ndims - 1];
    jpp.oh = (ndims == 3) ? 1 : dst_d.dims()[ndims - 2];

    jpp.ndims = ndims;
    jpp.mb = src_d.dims()[0];
    jpp.c_without_padding = src_d.dims()[1];
    switch (isa) {
        case sve_512: jpp.c_block = 16; break;
        default: jpp.c_block = 8; break;
    }

    jpp.alg = pd.alg_kind;
    jpp.tmp_md = memory_desc_t();

    using namespace format_tag;
    const auto blocked_fmt_tag = utils::one_of(isa, sve_512)
            ? utils::pick(ndims - 3, nCw16c, nChw16c, nCdhw16c)
            : utils::pick(ndims - 3, nCw8c, nChw8c, nCdhw8c);

    // src_d.data_type() is equal to dst_d.data_type(). This is checked in init
    auto ncsp_fmt_tag = format_tag::undef;

    const unsigned int L3_cache_size_per_core
            = platform::get_per_core_cache_size(3);
    const size_t block_size
            = ((size_t)jpp.id * jpp.ih * jpp.iw + jpp.od * jpp.oh * jpp.ow)
            * jpp.c_block * types::data_type_size(src_d.data_type());

    const bool forward_ncsp_allowed = !jpp.is_backward
            && jpp.c_without_padding > 3
            && ((jpp.ih > 1 && jpp.iw > 1
                        && block_size <= L3_cache_size_per_core)
                    || utils::one_of(src_d.data_type(), data_type::bf16,
                            data_type::f16));

    const bool backward_ncsp_allowed = jpp.is_backward
            && ((jpp.ih > 1 && jpp.iw > 1 && jpp.c_without_padding > 1
                        && block_size <= L3_cache_size_per_core)
                    || (utils::one_of(src_d.data_type(), data_type::bf16,
                                data_type::f16)
                            && !(jpp.alg == pooling_max
                                    && block_size > L3_cache_size_per_core)));

    ncsp_fmt_tag
            = ((forward_ncsp_allowed || backward_ncsp_allowed)
                      && utils::one_of(isa, sve_512, sve_256) && ndims <= 5)
            ? utils::pick(ndims - 3, ncw, nchw, ncdhw)
            : format_tag::undef;

    const auto nspc_fmt_tag = (ndims <= 5)
            ? utils::pick(ndims - 3, nwc, nhwc, ndhwc)
            : format_tag::undef;

    const auto fmt_tag = src_d.matches_one_of_tag(
            blocked_fmt_tag, ncsp_fmt_tag, nspc_fmt_tag);

    if (!dst_d.matches_tag(fmt_tag)) return status::unimplemented;

    if (!post_ops_ok(jpp, attr, dst_d)) return status::unimplemented;

    if (fmt_tag == ncsp_fmt_tag) {
        // transform input to blocked f32, call f32 jit, transform result to
        // plain output
        jpp.is_bf16 = false;
        jpp.is_f16 = false;
        jpp.dt_size = types::data_type_size(data_type::f32);
        jpp.tag_kind = jit_memory_tag_kind_t::ncsp;

        // used to initialize binary post-ops
        if (ppd->is_fwd() && jpp.with_binary) {
            CHECK(memory_desc_init_by_tag(jpp.tmp_md, ndims, dst_d.md_->dims,
                    data_type::f32, blocked_fmt_tag));
        }
    } else {
        jpp.is_bf16 = (src_d.data_type() == data_type::bf16
                && dst_d.data_type() == data_type::bf16);
        jpp.is_f16 = (src_d.data_type() == data_type::f16
                && dst_d.data_type() == data_type::f16);
        jpp.dt_size = types::data_type_size(src_d.data_type());
        jpp.tag_kind = (fmt_tag == nspc_fmt_tag)
                ? jit_memory_tag_kind_t::nspc
                : jit_memory_tag_kind_t::blocked;
    }

    if (ppd->is_fwd() && jpp.with_binary) {
        CHECK(set_binary_postops_formats(attr.post_ops_,
                jpp.tag_kind == jit_memory_tag_kind_t::ncsp ? &jpp.tmp_md
                                                            : dst_d.md_));
    }

    jpp.isa = isa;

    const bool args_ok = true && mayiuse(isa) && (fmt_tag != format_tag::undef)
            && jpp.is_bf16 == false && jpp.is_f16 == false
            && utils::one_of(pd.alg_kind, pooling_max,
                    pooling_avg_include_padding, pooling_avg_exclude_padding);
    if (!args_ok) return status::unimplemented;

    jpp.c = jpp.tag_kind == jit_memory_tag_kind_t::blocked
            ? utils::rnd_up(jpp.c_without_padding, jpp.c_block)
            : jpp.c_without_padding;
    if (jpp.tag_kind == jit_memory_tag_kind_t::blocked)
        assert(src_d.padded_dims()[1] == jpp.c);
    jpp.nb_c = utils::div_up(jpp.c, jpp.c_block);
    jpp.c_tail = jpp.c_without_padding % jpp.c_block;
    jpp.is_c_padded = jpp.tag_kind == jit_memory_tag_kind_t::blocked
            && src_d.padded_dims()[1] != jpp.c_without_padding;

    jpp.stride_d = (ndims == 5) ? pd.strides[0] : 1;
    jpp.stride_h = (ndims == 3) ? 1 : pd.strides[ndims - 4];
    jpp.stride_w = pd.strides[ndims - 3];
    jpp.kd = (ndims == 5) ? pd.kernel[0] : 1;
    jpp.kh = (ndims == 3) ? 1 : pd.kernel[ndims - 4];
    jpp.kw = pd.kernel[ndims - 3];

    jpp.f_pad = (ndims == 5) ? pd.padding[0][0] : 0;
    jpp.t_pad = (ndims == 3) ? 0 : pd.padding[0][ndims - 4];
    jpp.l_pad = pd.padding[0][ndims - 3];

    const int back_pad = calculate_end_padding(
            jpp.f_pad, jpp.od, jpp.id, jpp.stride_d, jpp.kd);
    const int bottom_pad = calculate_end_padding(
            jpp.t_pad, jpp.oh, jpp.ih, jpp.stride_h, jpp.kh);
    const int right_pad = calculate_end_padding(
            jpp.l_pad, jpp.ow, jpp.iw, jpp.stride_w, jpp.kw);

    if (jpp.f_pad >= jpp.kd || jpp.t_pad >= jpp.kh || jpp.l_pad >= jpp.kw
            || back_pad >= jpp.kd || bottom_pad >= jpp.kh
            || right_pad >= jpp.kw)
        return status::unimplemented;

    jpp.ind_dt = ppd->workspace_md() ? ppd->workspace_md()->data_type
                                     : data_type::undef;

    jpp.simple_alg = jpp.is_training
            || IMPLICATION(jpp.is_backward, jpp.kd <= jpp.stride_d);

    jpp.ur = 0;
    if (jpp.alg == pooling_max) {
        jpp.ur = 16;

        if (jpp.is_training)
            jpp.ur = 9;
        else if (jpp.is_backward)
            jpp.ur = 6;
    } else {
        if (jpp.is_backward)
            jpp.ur = 12;
        else
            jpp.ur = 24;
    }

    // select jpp.ur_bc
    if (jpp.tag_kind == jit_memory_tag_kind_t::nspc) {
        auto min_ur_w = nstl::max(1, utils::div_up(jpp.l_pad, jpp.stride_w));
        int min_ur_w1 = utils::div_up(nstl::max(0, right_pad), jpp.stride_w);
        if (min_ur_w < min_ur_w1) { min_ur_w = min_ur_w1; }
        jpp.ur_bc = nstl::min(jpp.nb_c, nstl::max(1, jpp.ur / min_ur_w));
        //take into account threading - to have enough work for parallelization
        float best_eff = 0;
        for (int ur_bc = jpp.ur_bc; ur_bc > 0; ur_bc--) {

            const auto nb2_c = utils::div_up(jpp.nb_c, ur_bc);
            auto work = jpp.is_backward
                    ? (ndims == 5 && jpp.simple_alg ? jpp.od : 1)
                    : (ndims == 5 ? jpp.od : jpp.oh);
            work *= jpp.mb * nb2_c;
            auto eff = (float)work / utils::rnd_up(work, jpp.nthr);
            if (eff > best_eff) {

                best_eff = eff;
                jpp.ur_bc = ur_bc;
            }
            if (eff > 0.9f) break; // Heuristic threshold
        }

        //take into account cache re-usage after zeroing on backward
        if (jpp.is_backward && ndims < 5) {
            const int L2 = platform::get_per_core_cache_size(2)
                    / sizeof(jpp.dt_size);
            int ur_bc = nstl::max(1, L2 / (jpp.kh * jpp.iw * jpp.c_block));
            jpp.ur_bc = nstl::min(jpp.ur_bc, ur_bc);
        }

        jpp.ur_bc_tail = jpp.nb_c % jpp.ur_bc;
    } else {
        jpp.ur_bc = 1;
        jpp.ur_bc_tail = 0;
    }

    // scratchpad for c_block slice of input and/or output
    using namespace memory_tracking::names;
    const int nscr = nstl::min(dnnl_get_max_threads(), jpp.mb * jpp.nb_c);
    if (jpp.tag_kind == jit_memory_tag_kind_t::ncsp) {
        scratchpad.book(key_pool_src_plain2blocked_cvt,
                static_cast<size_t>(jpp.c_block) * jpp.id * jpp.ih * jpp.iw
                        * nscr,
                jpp.dt_size);
        scratchpad.book(key_pool_dst_plain2blocked_cvt,
                static_cast<size_t>(jpp.c_block) * jpp.od * jpp.oh * jpp.ow
                        * nscr,
                jpp.dt_size);
        scratchpad.book<uint32_t>(key_pool_ind_plain2blocked_cvt,
                static_cast<size_t>(jpp.c_block) * jpp.od * jpp.oh * jpp.ow
                        * nscr);
    }

    jpp.post_ops = attr.post_ops_;

    return status::success;
}

static int reg_ind(int shift, int bc, int j, int ur_bc, int ur_w) noexcept {
    return shift * ur_bc * ur_w + bc * ur_w + j;
}

template <cpu_isa_t isa>
inline void jit_uni_pool_kernel_t<isa>::prepare_tail_mask() {
    /* PRegS(k_c_tail_mask) keeps flags in the context
           of 8-bit elements. */
    set_preg(k_c_tail_mask_b.b, jpp.c_tail, X_TMP_0, X_TMP_1);
    set_preg(k_c_tail_mask_s.s, jpp.c_tail, X_TMP_0, X_TMP_1);
    not_(k_c_tail_mask_s_not.b, P_ALL_ONE / T_z, k_c_tail_mask_s.b);
}

template <cpu_isa_t isa>
inline void jit_uni_pool_kernel_t<isa>::load(const int idx,
        const xreg_t &reg_ptr, const int offset,
        const bool is_c_tail_proccessing) {
    if (is_c_tail_proccessing && !jpp.is_c_padded) {
        add_imm(X_DEFAULT_ADDR, reg_ptr, offset, X_TMP_0);
        ld1w(ZRegS(idx), k_c_tail_mask_s / T_z, ptr(X_DEFAULT_ADDR));
    } else {
        add_imm(X_DEFAULT_ADDR, reg_ptr, offset, X_TMP_0);
        ld1w(ZRegS(idx), P_ALL_ONE / T_z, ptr(X_DEFAULT_ADDR));
    }
}

template <cpu_isa_t isa>
inline void jit_uni_pool_kernel_t<isa>::store(const int idx,
        const xreg_t &reg_ptr, const int offset,
        const bool is_c_tail_proccessing) {
    add_imm(X_DEFAULT_ADDR, reg_ptr, offset, X_TMP_0);

    if (is_c_tail_proccessing) {
        if (!jpp.is_c_padded) {
            st1w(ZRegS(idx), k_c_tail_mask_s, ptr(X_DEFAULT_ADDR));
        } else {
            if (jpp.with_postops) mov(ZRegS(idx), k_c_tail_mask_s_not / T_m, 0);
            st1w(ZRegS(idx), P_ALL_ONE, ptr(X_DEFAULT_ADDR));
        }
    } else {
        st1w(ZRegS(idx), P_ALL_ONE, ptr(X_DEFAULT_ADDR));
    }
}

template <cpu_isa_t isa>
bool jit_uni_pool_kernel_t<isa>::post_ops_ok(jit_pool_conf_t &jpp,
        const primitive_attr_t &attr, const memory_desc_wrapper &dst_d) {
    const auto &post_ops = attr.post_ops_;
    const auto &entries = post_ops.entry_;
    jpp.with_postops = false;
    jpp.with_eltwise = false;
    jpp.with_binary = false;

    if (!jpp.is_backward) {
        for (const auto &entry : entries) {
            if (entry.is_eltwise()) {
                const auto alg = entry.eltwise.alg;
                jpp.with_eltwise
                        = eltwise_injector::is_supported(to_vla_sve(isa), alg);
            } else if (entry.is_binary()) {
                if (entry.binary.src1_desc.data_type == data_type::bf16
                        || entry.binary.src1_desc.data_type == data_type::f16)
                    return false;

                jpp.with_binary = true;
            } else
                return false;
        }

        jpp.with_postops = jpp.with_eltwise || jpp.with_binary;
    }

    return binary_injector::binary_args_broadcast_supported(
            post_ops, dst_d, get_supported_bcast_strategies());
}

template <cpu_isa_t isa>
void jit_uni_pool_kernel_t<isa>::apply_postops(int ur_bc, int ur_w, int c_block,
        const std::function<bool(int)> &is_tail_predicate) {
    binary_injector::rhs_arg_dynamic_params_t rhs_arg_params;
    const int end_idx = vmm_idx_upper_bound() + 1;
    const int start_idx = end_idx - (ur_bc * ur_w);

    if (jpp.with_binary) {

        const int c_off = (jpp.tag_kind == jit_memory_tag_kind_t::nspc)
                ? jpp.c
                : c_block;

        if (jpp.tag_kind == jit_memory_tag_kind_t::ncsp) {
            ldr(X_TMP_0, ptr(reg_param, GET_OFF(dst)));
            ldr(X_TMP_1, ptr(reg_param, GET_OFF(dst_po_helper)));
            sub(tmp_gpr, reg_output, X_TMP_0);
            add(tmp_gpr, tmp_gpr, X_TMP_1);
        }

        for (int jj = 0; jj < ur_w; jj++) {
            for (int bci = 0; bci < ur_bc; bci++) {
                const auto vmm_idx
                        = vreg(reg_ind(0, bci, jj, ur_bc, ur_w)).getIdx();

                const size_t output_offset
                        = jpp.dt_size * (jj * c_off + bci * c_block);

                rhs_arg_params.vmm_idx_to_out_reg.emplace(vmm_idx,
                        jpp.tag_kind == jit_memory_tag_kind_t::ncsp
                                ? tmp_gpr
                                : reg_output);
                rhs_arg_params.vmm_idx_to_out_elem_off_val.emplace(
                        vmm_idx, output_offset);
                if (is_tail_predicate && is_tail_predicate(bci)) {
                    rhs_arg_params.vmm_tail_idx_.emplace(vmm_idx);
                }
            }
        }
    }
    postops_injector_->compute_vector_range(start_idx, end_idx, rhs_arg_params);
}

template <cpu_isa_t isa>
inline void jit_uni_pool_kernel_t<isa>::maybe_recalculate_divisor(
        int jj, int ur_w, int pad_l, int pad_r, bool with_c_tail_proccessing) {
    if (jpp.alg == pooling_avg_exclude_padding) {
        int kw = jpp.kw;
        int stride_w = jpp.stride_w;

        int non_zero_kw = kw;
        non_zero_kw -= nstl::max(0, pad_l - jj * stride_w);
        non_zero_kw -= nstl::max(0, pad_r - (ur_w - 1 - jj) * stride_w);

        if (non_zero_kw != prev_kw) {
            mov_imm(W_TMP_0, float2int((float)non_zero_kw));
            dup(vmm_tmp.s, W_TMP_0);

            fmul(vmm_tmp.s, vmm_tmp.s, vmm_ker_area_h);
            prev_kw = non_zero_kw;
        }
    }
}

template <cpu_isa_t isa>
inline void jit_uni_pool_kernel_t<isa>::avg_step(int ur_w, int ur_bc, int pad_l,
        int pad_r, bool with_c_tail_proccessing) {

    auto iw = jpp.iw;
    auto kw = jpp.kw;
    auto stride_w = jpp.stride_w;
    auto c_block = jpp.c_block;
    auto dt_size = jpp.dt_size;
    const int c_off
            = (jpp.tag_kind == jit_memory_tag_kind_t::nspc) ? jpp.c : c_block;
    Label kd_label, kh_label;

    const auto is_tail_processing = [&](int bc) {
        return with_c_tail_proccessing && bc == (ur_bc - 1);
    };

    for (int jj = 0; jj < ur_w; jj++) {
        if (jpp.is_backward)
            maybe_recalculate_divisor(
                    jj, ur_w, pad_l, pad_r, with_c_tail_proccessing);
        for (int bci = 0; bci < ur_bc; bci++) {
            const auto accr_i = reg_ind(0, bci, jj, ur_bc, ur_w);
            auto accvr = vreg(accr_i);
            if (jpp.is_backward) {
                auto output_offset = dt_size * (jj * c_off + bci * c_block);
                load(accvr.getIdx(), reg_output, output_offset,
                        is_tail_processing(bci));
                fdiv(accvr.s, P_ALL_ONE, vmm_tmp.s);
            } else {
                eor(accvr.d, accvr.d, accvr.d);
            }
        }
    }

    if (jpp.simple_alg && jpp.ndims == 5) {
        str(reg_input, pre_ptr(X_TRANSLATOR_STACK, -8));
        str(reg_output, pre_ptr(X_TRANSLATOR_STACK, -8));

        mov(aux_reg_input_d, reg_input);
        ldr(ki, ptr(reg_param, GET_OFF(kd_padding)));
        L(kd_label);
        mov(aux_reg_input, aux_reg_input_d);
    } else {
        mov(aux_reg_input, reg_input);
    }

    eor(kj, kj, kj);
    L(kh_label);
    {
        for (int ki = 0; ki < kw; ki++) {
            int jj_start = utils::div_up(nstl::max(0, pad_l - ki), stride_w);
            int jj_end = ur_w
                    - utils::div_up(
                            nstl::max(0, ki + pad_r - (kw - 1)), stride_w);

            for_(int jj = jj_start; jj < jj_end; jj++)
            for (int bci = 0; bci < ur_bc; bci++) {
                const auto accvr = vreg(reg_ind(0, bci, jj, ur_bc, ur_w)).s;
                const auto inpr_i = reg_ind(1, bci, jj, ur_bc, ur_w);
                auto inpvr = vreg(inpr_i);
                int aux_input_offset
                        = (ki + jj * stride_w - pad_l) * c_off + bci * c_block;
                if (aux_input_offset >= iw * c_off) continue;
                int input_offset = dt_size * aux_input_offset;
                if (jpp.is_backward) {
                    load(reg_idx(inpr_i), aux_reg_input, input_offset,
                            is_tail_processing(bci));

                    fadd(inpvr.s, inpvr.s, accvr);
                    store(reg_idx(inpr_i), aux_reg_input, input_offset,
                            is_tail_processing(bci));
                } else {
                    if (is_tail_processing(bci)) {
                        load(vmm_tmp_1.getIdx(), aux_reg_input, input_offset,
                                is_tail_processing(bci));
                        fadd(accvr, accvr, vmm_tmp_1);
                    } else {
                        add_imm(X_DEFAULT_ADDR, aux_reg_input, input_offset,
                                X_TMP_0);
                        ld1w(z_tmp0.s, P_ALL_ONE / T_z, ptr(X_DEFAULT_ADDR));
                        fadd(accvr, accvr, z_tmp0.s);
                    }
                }
            }
        }
        add_imm(aux_reg_input, aux_reg_input, (jpp.dt_size * iw * c_off),
                X_TMP_0);
        adds(kj, kj, 1);
        cmp(kj, reg_kh);
        b(LT, kh_label);
    }

    if (jpp.simple_alg && jpp.ndims == 5) {
        add_imm(aux_reg_input_d, aux_reg_input_d,
                (jpp.dt_size * jpp.ih * iw * c_off), X_TMP_0);
        subs(ki, ki, 1);
        cmp(ki, 0);
        b(GT, kd_label);
        ldr(reg_output, post_ptr(X_TRANSLATOR_STACK, 8));
        ldr(reg_input, post_ptr(X_TRANSLATOR_STACK, 8));
    }

    if (!jpp.is_backward) {
        for (int jj = 0; jj < ur_w; jj++) {
            maybe_recalculate_divisor(
                    jj, ur_w, pad_l, pad_r, with_c_tail_proccessing);
            for (int bci = 0; bci < ur_bc; bci++) {
                const auto accr_i = reg_ind(0, bci, jj, ur_bc, ur_w);
                const auto accvr = vreg(accr_i);
                fdiv(accvr.s, P_ALL_ONE, vmm_tmp.s);
            }
        }

        if (jpp.with_postops)
            apply_postops(ur_bc, ur_w, c_block, is_tail_processing);

        for (int jj = 0; jj < ur_w; jj++) {
            for (int bci = 0; bci < ur_bc; bci++) {
                const auto accr_i = reg_ind(0, bci, jj, ur_bc, ur_w);
                const auto output_offset
                        = dt_size * (jj * c_off + bci * c_block);
                store(reg_idx(accr_i), reg_output, output_offset,
                        is_tail_processing(bci));
            }
        }
    }
}

template <cpu_isa_t isa>
inline void jit_uni_pool_kernel_t<isa>::max_step_fwd(int ur_w, int ur_bc,
        int pad_l, int pad_r, bool with_c_tail_proccessing) {
    int iw = jpp.iw;
    int kw = jpp.kw;
    int stride_w = jpp.stride_w;
    int c_block = jpp.c_block;
    const int c_off
            = (jpp.tag_kind == jit_memory_tag_kind_t::nspc) ? jpp.c : c_block;
    Label kd_label, kh_label;

    auto is_tail_processing = [&](int bc) {
        return with_c_tail_proccessing && bc == (ur_bc - 1);
    };

    mov_imm(W_TMP_0, float2int(nstl::numeric_limits<float>::lowest()));
    dup(vmm_tmp.s, W_TMP_0);

    for_(int jj = 0; jj < ur_w; jj++)
    for (int bci = 0; bci < ur_bc; bci++) {
        const auto accvr = vreg(reg_ind(0, bci, jj, ur_bc, ur_w)).d;
        mov(accvr, vmm_tmp.d);

        if (jpp.is_training) {
            const auto indvr = vreg(reg_ind(2, bci, jj, ur_bc, ur_w)).d;

            eor(indvr, indvr, indvr);
        }
    }
    if (jpp.is_training) dup(vmm_k_offset, reg_k_shift);
    if (jpp.ndims == 5) {
        str(reg_input, pre_ptr(X_TRANSLATOR_STACK, -8));

        str(reg_output, pre_ptr(X_TRANSLATOR_STACK, -8));
        mov(aux_reg_input_d, reg_input);
        ldr(ki, ptr(reg_param, GET_OFF(kd_padding)));
        L(kd_label);
        mov(aux_reg_input, aux_reg_input_d);
    } else {
        mov(aux_reg_input, reg_input);
    }
    eor(kj, kj, kj);
    L(kh_label);
    {
        for (int ki = 0; ki < kw; ki++) {
            int jj_start = utils::div_up(nstl::max(0, pad_l - ki), stride_w);
            int jj_end = ur_w
                    - utils::div_up(
                            nstl::max(0, ki + pad_r - (kw - 1)), stride_w);
            for_(int jj = jj_start; jj < jj_end; jj++)
            for (int bci = 0; bci < ur_bc; bci++) {
                const auto accvr = vreg(reg_ind(0, bci, jj, ur_bc, ur_w)).s;
                const auto inpr_i = reg_ind(1, bci, jj, ur_bc, ur_w);
                const auto inpvr = vreg(inpr_i).s;
                const auto indvr = vreg(reg_ind(2, bci, jj, ur_bc, ur_w)).s;
                int aux_input_offset
                        = (ki + jj * stride_w - pad_l) * c_off + bci * c_block;
                if (aux_input_offset >= iw * c_off) continue;
                int input_offset = jpp.dt_size * aux_input_offset;
                load(reg_idx(inpr_i), aux_reg_input, input_offset,
                        is_tail_processing(bci));

                fcmlt(k_store_mask.s, P_ALL_ONE / T_z, accvr, inpvr);
                sel(accvr, k_store_mask / T_m, inpvr, accvr);
                if (jpp.is_training) {
                    sel(indvr, k_store_mask / T_m, vmm_k_offset, indvr);
                }
            }
            if (jpp.is_training) add(vmm_k_offset, vmm_k_offset, vmm_one);
        }

        add_imm(aux_reg_input, aux_reg_input, (jpp.dt_size * iw * c_off),
                X_TMP_0);
        adds(kj, kj, 1);
        cmp(kj, reg_kh);
        b(LT, kh_label);
    }

    if (jpp.ndims == 5) {
        add_imm(aux_reg_input_d, aux_reg_input_d,
                (jpp.dt_size * jpp.ih * iw * c_off), X_TMP_0);
        if (jpp.is_training) {
            ldr(tmp_gpr, ptr(reg_param, GET_OFF(kd_padding_shift)));
            dup(vmm_tmp.s, WReg(tmp_gpr.getIdx()));
            add(vmm_k_offset, vmm_k_offset, vmm_tmp.s);
        }

        subs(ki, ki, 1);
        cmp(ki, 0);
        b(GT, kd_label);
        ldr(reg_output, post_ptr(X_TRANSLATOR_STACK, 8));
        ldr(reg_input, post_ptr(X_TRANSLATOR_STACK, 8));
    }

    if (jpp.with_postops)
        apply_postops(ur_bc, ur_w, c_block, is_tail_processing);

    for_(int jj = 0; jj < ur_w; jj++)
    for (int bci = 0; bci < ur_bc; bci++) {
        const auto accr_i = reg_ind(0, bci, jj, ur_bc, ur_w);
        const auto output_offset = jpp.dt_size * (jj * c_off + bci * c_block);
        store(reg_idx(accr_i), reg_output, output_offset,
                is_tail_processing(bci));

        if (jpp.is_training) {
            const size_t step_index = (jj * c_off + bci * c_block)
                    * types::data_type_size(jpp.ind_dt);

            const auto indr_i = reg_ind(2, bci, jj, ur_bc, ur_w);
            auto vr = vreg(indr_i);
            if (jpp.ind_dt == data_type::u8) {
                add_imm(X_DEFAULT_ADDR, reg_index, step_index, X_TMP_0);
                if (is_tail_processing(bci)) {
                    if (jpp.is_c_padded) {
                        mov(vr.s, k_c_tail_mask_s_not / T_m, 0);
                        umin(vr.s, 255);
                        st1b(vr.s, P_ALL_ONE, ptr(X_DEFAULT_ADDR));
                    } else {
                        umin(vr.s, 255);
                        st1b(vr.s, k_c_tail_mask_s, ptr(X_DEFAULT_ADDR));
                    }
                } else {
                    umin(vr.s, 255);
                    st1b(vr.s, P_ALL_ONE, ptr(X_DEFAULT_ADDR));
                }
            } else {
                store(vr.getIdx(), reg_index, step_index,
                        is_tail_processing(bci));
            }
        }
    }
}

template <cpu_isa_t isa>
inline void jit_uni_pool_kernel_t<isa>::max_step_bwd(int ur_w, int ur_bc,
        int pad_l, int pad_r, bool with_c_tail_proccessing) {

    int iw = jpp.iw;
    int kw = jpp.kw;
    int stride_w = jpp.stride_w;
    int c_block = jpp.c_block;
    const int c_off
            = (jpp.tag_kind == jit_memory_tag_kind_t::nspc) ? jpp.c : c_block;
    Label kd_label, kh_label;

    const auto is_tail_processing = [&](int bc) {
        return with_c_tail_proccessing && bc == (ur_bc - 1);
    };

    for_(int jj = 0; jj < ur_w; jj++)
    for (int bci = 0; bci < ur_bc; bci++) {
        const auto outr_i = reg_ind(0, bci, jj, ur_bc, ur_w);
        auto out_offset = jpp.dt_size * (jj * c_off + bci * c_block);
        load(reg_idx(outr_i), reg_output, out_offset, is_tail_processing(bci));
        const size_t step_index = (jj * c_off + bci * c_block)
                * types::data_type_size(jpp.ind_dt);

        const auto indr_i = reg_ind(1, bci, jj, ur_bc, ur_w);
        auto indvr = vreg(indr_i);
        if (jpp.ind_dt == data_type::u8) {
            if (is_tail_processing(bci) && !jpp.is_c_padded) {
                add_imm(X_DEFAULT_ADDR, reg_index, step_index, X_TMP_0);
                ld1b(indvr.s, k_c_tail_mask_s / T_z, ptr(X_DEFAULT_ADDR));
            } else {
                add_imm(X_DEFAULT_ADDR, reg_index, step_index, X_TMP_0);
                ld1b(indvr.s, P_ALL_ONE / T_z, ptr(X_DEFAULT_ADDR));
            }
        } else {
            load(indvr.getIdx(), reg_index, step_index,
                    is_tail_processing(bci));
        }
    }
    dup(vmm_k_offset, reg_k_shift);

    if (jpp.simple_alg && jpp.ndims == 5) {
        str(reg_input, pre_ptr(X_TRANSLATOR_STACK, -8));
        str(reg_output, pre_ptr(X_TRANSLATOR_STACK, -8));
        mov(aux_reg_input_d, reg_input);

        ldr(ki, ptr(reg_param, GET_OFF(kd_padding)));
        ldr(reg_kd_pad_shift, ptr(reg_param, GET_OFF(kd_padding_shift)));
        L(kd_label);
        mov(aux_reg_input, aux_reg_input_d);
    } else {
        mov(aux_reg_input, reg_input);
    }

    eor(kj, kj, kj);
    L(kh_label);
    {
        for (int ki = 0; ki < kw; ki++) {
            int jj_start = utils::div_up(nstl::max(0, pad_l - ki), stride_w);
            int jj_end = ur_w
                    - utils::div_up(
                            nstl::max(0, ki + pad_r - (kw - 1)), stride_w);
            for_(int jj = jj_start; jj < jj_end; jj++)
            for (int bci = 0; bci < ur_bc; bci++) {
                const auto outvr = vreg(reg_ind(0, bci, jj, ur_bc, ur_w));
                const auto indvr = vreg(reg_ind(1, bci, jj, ur_bc, ur_w)).s;
                const auto inpr_i = reg_ind(2, bci, jj, ur_bc, ur_w);
                const auto inpvr = vreg(inpr_i).s;
                int aux_inp_offset
                        = (ki + jj * stride_w - pad_l) * c_off + bci * c_block;
                if (aux_inp_offset >= iw * c_off) continue;
                int inp_offset = jpp.dt_size * aux_inp_offset;
                load(reg_idx(inpr_i), aux_reg_input, inp_offset,
                        is_tail_processing(bci));
                cmpeq(k_store_mask.s, P_ALL_ONE / T_z, indvr, vmm_k_offset);
                fadd(inpvr, k_store_mask / T_m, outvr.s);
                store(inpvr.getIdx(), aux_reg_input, inp_offset,
                        is_tail_processing(bci));
            }

            add(vmm_k_offset, vmm_k_offset, vmm_one);
        }
        add_imm(aux_reg_input, aux_reg_input, (jpp.dt_size * iw * c_off),
                X_TMP_0);
        adds(kj, kj, 1);
        cmp(kj, reg_kh);
        b(LT, kh_label);
    }
    if (jpp.simple_alg && jpp.ndims == 5) {
        add_imm(aux_reg_input_d, aux_reg_input_d,
                (jpp.dt_size * jpp.ih * iw * c_off), X_TMP_0);

        dup(vmm_tmp.s, WReg(reg_kd_pad_shift.getIdx()));

        add(vmm_k_offset, vmm_k_offset, vmm_tmp.s);
        subs(ki, ki, 1);
        cmp(ki, 0);
        b(GT, kd_label);
        ldr(reg_output, post_ptr(X_TRANSLATOR_STACK, 8));
        ldr(reg_input, post_ptr(X_TRANSLATOR_STACK, 8));
    }
}

template <cpu_isa_t isa>
void jit_uni_pool_kernel_t<isa>::zero_diff_src(
        int ur_bc, bool with_c_tail_proccessing) {
    const int c_off = (jpp.tag_kind == jit_memory_tag_kind_t::nspc)
            ? jpp.c
            : jpp.c_block;

    Label l_skip, l_ih_loop, l_id_loop;

    auto is_tail_processing = [&](int bc) {
        return with_c_tail_proccessing && bc == (ur_bc - 1);
    };

    ldr(reg_zero_id, ptr(reg_param, GET_OFF(zero_id)));
    cmp(reg_zero_id, 0);
    b(EQ, l_skip);

    ldr(reg_zero_ih, ptr(reg_param, GET_OFF(zero_ih)));
    cmp(reg_zero_ih, 0);
    b(EQ, l_skip);

    ldr(reg_zero_ptr, ptr(reg_param, GET_OFF(zero_ptr)));

    TReg vzero = vmm_tmp;
    eor(vzero.d, vzero.d, vzero.d);

    const int width_size = jpp.iw * c_off * jpp.dt_size;

    auto aux_reg_zero_ptr = tmp_gpr;

    L(l_id_loop);
    {
        mov(aux_reg_zero_ptr, reg_zero_ptr);
        mov(aux_reg_zero_ih, reg_zero_ih);
        L(l_ih_loop);
        {
            const int step = c_off * jpp.dt_size;

            // TODO: maybe a big code generated here
            for_(int i = 0; i < width_size; i += step)
            for (int bci = 0; bci < ur_bc; bci++) {
                const int offs = i + bci * jpp.c_block * jpp.dt_size;
                store(vzero.getIdx(), reg_zero_ptr, offs,
                        is_tail_processing(bci));
            }
            add_imm(reg_zero_ptr, reg_zero_ptr, width_size, X_TMP_0);
            subs(aux_reg_zero_ih, aux_reg_zero_ih, 1);
            b(NE, l_ih_loop);
        }
        mov(reg_zero_ptr, aux_reg_zero_ptr);
        add_imm(reg_zero_ptr, reg_zero_ptr, (width_size * jpp.ih), X_TMP_0);
        subs(reg_zero_id, reg_zero_id, 1);
        b(NE, l_id_loop);
    }

    L(l_skip);
}

template <cpu_isa_t isa>
void jit_uni_pool_kernel_t<isa>::generate() {

    this->preamble();

    size_t simd_w_ = cpu_isa_traits<isa>::vlen / sizeof(float);

    if (simd_w_ != cpu_sveLen / sizeof(float))
        set_preg(P_ALL_ONE.s, simd_w_, X_TMP_0, X_TMP_1);

    Label idx_table;

    int ow = jpp.ow;
    int iw = jpp.iw;
    int kw = jpp.kw;
    int kh = jpp.kh;
    int c_block = jpp.c_block;
    int stride_w = jpp.stride_w;
    int l_pad = jpp.l_pad;
    const int c_off
            = (jpp.tag_kind == jit_memory_tag_kind_t::nspc) ? jpp.c : c_block;

    pfalse(p_all_zero.b);

    ldr(reg_input, ptr(reg_param, GET_OFF(src)));
    ldr(reg_output, ptr(reg_param, GET_OFF(dst)));
    if (jpp.alg == pooling_max && (jpp.is_training || jpp.is_backward))
        ldr(reg_index, ptr(reg_param, GET_OFF(indices)));
    ldr(reg_kh, ptr(reg_param, GET_OFF(kh_padding)));
    ldr(reg_k_shift, ptr(reg_param, GET_OFF(kh_padding_shift)));
    ldr(reg_ker_area_h, ptr(reg_param, GET_OFF(ker_area_h)));
    ldr(reg_nbc, ptr(reg_param, GET_OFF(ur_bc)));

    auto process_oi
            = [&](int ur_w, int ur_bc, int lpad, int rpad,
                      bool with_c_tail_proccessing, bool inc_reg = true) {
        step(ur_w, ur_bc, lpad, rpad, with_c_tail_proccessing);

        if (!inc_reg) return;

        auto dt_size = jpp.dt_size;
        add_imm(reg_input, reg_input,
                (dt_size * nstl::max(0, ur_w * stride_w - lpad) * c_off),
                X_TMP_0);
        add_imm(reg_output, reg_output, (dt_size * ur_w * c_off), X_TMP_0);

        if (jpp.alg == pooling_max && (jpp.is_training || jpp.is_backward)) {
            auto ind_dt_size = types::data_type_size(jpp.ind_dt);
            add_imm(reg_index, reg_index, ((ur_w * c_off) * ind_dt_size),
                    X_TMP_0);
        }
    };

    auto perform_ker = [&](int ur_bc, bool with_c_tail_processing) {
        prev_kw = 0; // re-initialize this value for avg steps

        if (jpp.is_backward && jpp.simple_alg)
            zero_diff_src(ur_bc, with_c_tail_processing);

        if (jpp.alg == pooling_avg_exclude_padding) {
            // vmm_ker_area_h and vmm_c_tail_mask are stored in one register
            // so when vmm_c_tail_mask is used we need to load vmm_ker_area_h
            // exactly where this information is needed with the
            // vmm_c_tail_mask information being saved first
            dup(TReg(vmm_ker_area_h.getIdx()).s, WReg(reg_ker_area_h.getIdx()));
        }

        if (jpp.alg == pooling_avg_include_padding) {
            mov_imm(W_TMP_0, float2int((float)(kw * kh * jpp.kd)));
            dup(vmm_tmp.s, W_TMP_0);
        }

        if (jpp.alg == pooling_max && (jpp.is_training || jpp.is_backward)) {
            // The same situation as above(vmm_ker_area_h).
            dup(vmm_one, 1);
        }

        const int ur_w = nstl::min(jpp.ow, jpp.ur / jpp.ur_bc);
        const int n_oi_iterations = utils::div_up(ow, ur_w);
        const int ur_stride_w = ur_w * stride_w;
        const int l_pad_iterations
                = nstl::min(n_oi_iterations, utils::div_up(l_pad, ur_stride_w));

        for (int i = 0; i < l_pad_iterations; ++i) {
            const int ow_s = i * ur_w;
            const int ow_e = nstl::min(ow, ow_s + ur_w);
            const int cur_l_pad = l_pad - i * ur_stride_w;
            const int cur_r_pad = nstl::max(
                    0, calculate_end_padding(l_pad, ow_e, iw, stride_w, kw));
            const int cur_ur_w = ow_e - ow_s;
            process_oi(cur_ur_w, ur_bc, cur_l_pad, cur_r_pad,
                    with_c_tail_processing);
        }

        const int rem_n_oi_iters = n_oi_iterations - l_pad_iterations;
        const int cur_iw = l_pad_iterations * ur_stride_w - l_pad;
        const int cur_iw_rightmost_idx = cur_iw + kw - 1;
        const int no_pad_full_n_oi_iters = utils::saturate<int>(
                0, rem_n_oi_iters, (iw - cur_iw_rightmost_idx) / ur_stride_w);

        if (no_pad_full_n_oi_iters > 0) {
            Label ow_loop;
            if (no_pad_full_n_oi_iters > 1) eor(oi_iter, oi_iter, oi_iter);
            L(ow_loop);
            {
                process_oi(ur_w, ur_bc, 0, 0, with_c_tail_processing);
                if (no_pad_full_n_oi_iters > 1) {
                    adds(oi_iter, oi_iter, 1);
                    cmp(oi_iter, no_pad_full_n_oi_iters);
                    b(LT, ow_loop);
                }
            }
        }

        for (int i = l_pad_iterations + no_pad_full_n_oi_iters;
                i < n_oi_iterations; ++i) {
            const int ow_s = i * ur_w;
            const int ow_e = nstl::min(ow, ow_s + ur_w);
            const int cur_r_pad = nstl::max(
                    0, calculate_end_padding(l_pad, ow_e, iw, stride_w, kw));
            const int cur_ur_w = ow_e - ow_s;
            process_oi(cur_ur_w, ur_bc, 0, cur_r_pad, with_c_tail_processing);
        }
    };
    Label ur_bc_tail_label, c_tail_processing_label, finish_label;

    if (jpp.ur_bc_tail > 0) {
        cmp(reg_nbc, jpp.ur_bc);
        b(NE, ur_bc_tail_label);
    } else if (jpp.c_tail != 0) {
        // ur_bc contains number of channel blocks to processing
        // b_c contains number of channel blocks already processed
        // If reg_nbc + tmp_gpr == jpp.nb_c then this is
        // information that probably channel tail processing will be needed.
        /* get mem address */
        ldr(tmp_gpr, ptr(reg_param, GET_OFF(b_c)));
        add(tmp_gpr, tmp_gpr, reg_nbc);
        cmp(tmp_gpr, jpp.nb_c);
        b(Xbyak_aarch64::EQ, c_tail_processing_label);
    }

    perform_ker(jpp.ur_bc, false);

    if (jpp.ur_bc_tail > 0) {
        bl(finish_label);

        // If ur_bc_tail exists then we know that this is
        // last set of blocks to process and we need
        // care of c tail processing if number of channels
        // is not divided by number of channels in block
        L(ur_bc_tail_label);
        if (jpp.c_tail != 0) prepare_tail_mask();
        perform_ker(jpp.ur_bc_tail, jpp.c_tail != 0);

        L(finish_label);
    } else if (jpp.c_tail != 0) {
        bl(finish_label);

        L(c_tail_processing_label);
        prepare_tail_mask();
        perform_ker(jpp.ur_bc, true);

        L(finish_label);
    }

    this->postamble();

    if (jpp.with_eltwise && postops_injector_)
        postops_injector_->prepare_table();
}

template struct jit_uni_pool_kernel_t<sve_512>;
template struct jit_uni_pool_kernel_t<sve_256>;

} // namespace aarch64
} // namespace cpu
} // namespace impl
} // namespace dnnl

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