path: root/src/core/NEON/kernels/arm_conv/pooling/kernels/a64_u8q_nhwc_max_generic_depthfirst/generic.cpp

/*
 * Copyright (c) 2021-2023 Arm Limited.
 *
 * SPDX-License-Identifier: MIT
 *
 * 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.
 */

#include "pooling.hpp"
#include <cstdint>
#include <cstddef>

#if defined(__aarch64__)

namespace arm_conv {
namespace pooling {


void a64_u8q_nhwc_max_generic_depthfirst_impl(
  const uint64_t,
  const uint64_t n_valid_cells,
  uint64_t n_channels,
  const uint8_t *const *const inptrs,
  uint8_t *outptr,
  const Requantize32 &qp
)
{
  __asm__ __volatile__(
    "cmp %x[n_channels], #0x40\n"
    "mov x27, #0x0\n"
    "mov x26, #0x10\n"  // cntb _, ALL, #1
    "mov x24, #0x20\n"  // cntb _, ALL, #2
    "mov x23, #0x30\n"  // cntb _, ALL, #3
    "blt 7f\n"
    "1:"  // 4-vectors of channels
    "lsr x25, %x[n_valid_cells], #0x2\n"
    "movi v8.16b, #0x0\n"
    "movi v7.16b, #0x0\n"
    "mov x22, %x[inptrs]\n"
    "movi v6.16b, #0x0\n"
    "movi v5.16b, #0x0\n"
    "cbz x25, 4f\n"
    "ldp x21, x20, [x22, #0x0]\n"
    "ldr q4, [x21, x27]\n"
    "subs x25, x25, #0x1\n"
    "ldr q3, [x20, x27]\n"
    "ldr q2, [x21, x26]\n"
    "ldr q1, [x20, x26]\n"
    "ldr q0, [x21, x24]\n"
    "ldr q31, [x20, x24]\n"
    "ldr q30, [x21, x23]\n"
    "ldr q29, [x20, x23]\n"
    "ldp x21, x20, [x22, #0x10]\n"
    "add x22, x22, #0x20\n"
    "ldr q28, [x21, x27]\n"
    "ldr q22, [x20, x27]\n"
    "ldr q27, [x21, x26]\n"
    "ldr q21, [x20, x26]\n"
    "ldr q26, [x21, x24]\n"
    "ldr q20, [x20, x24]\n"
    "ldr q25, [x21, x23]\n"
    "ldr q24, [x20, x23]\n"
    "beq 3f\n"
    "2:"  // 4-vectors of channels: 4 inputs loop
    "umax v23.16b, v4.16b, v3.16b\n"
    "umax v19.16b, v28.16b, v22.16b\n"
    "ldp x21, x20, [x22, #0x0]\n"
    "ldr q4, [x21, x27]\n"
    "ldr q3, [x20, x27]\n"
    "umax v22.16b, v2.16b, v1.16b\n"
    "ldr q2, [x21, x26]\n"
    "umax v18.16b, v27.16b, v21.16b\n"
    "ldr q1, [x20, x26]\n"
    "umax v21.16b, v0.16b, v31.16b\n"
    "ldr q0, [x21, x24]\n"
    "umax v17.16b, v26.16b, v20.16b\n"
    "ldr q31, [x20, x24]\n"
    "umax v20.16b, v30.16b, v29.16b\n"
    "ldr q30, [x21, x23]\n"
    "umax v16.16b, v25.16b, v24.16b\n"
    "ldr q29, [x20, x23]\n"
    "umax v19.16b, v23.16b, v19.16b\n"
    "umax v18.16b, v22.16b, v18.16b\n"
    "ldp x21, x20, [x22, #0x10]\n"
    "ldr q28, [x21, x27]\n"
    "ldr q22, [x20, x27]\n"
    "umax v17.16b, v21.16b, v17.16b\n"
    "umax v16.16b, v20.16b, v16.16b\n"
    "ldr q27, [x21, x26]\n"
    "ldr q21, [x20, x26]\n"
    "subs x25, x25, #0x1\n"
    "umax v8.16b, v8.16b, v19.16b\n"
    "ldr q26, [x21, x24]\n"
    "ldr q20, [x20, x24]\n"
    "umax v7.16b, v7.16b, v18.16b\n"
    "umax v6.16b, v6.16b, v17.16b\n"
    "ldr q25, [x21, x23]\n"
    "ldr q24, [x20, x23]\n"
    "umax v5.16b, v5.16b, v16.16b\n"
    "add x22, x22, #0x20\n"
    "bgt 2b\n"
    "3:"  // 4-vectors of channels: 4 inputs tail
    "umax v23.16b, v4.16b, v3.16b\n"
    "umax v19.16b, v28.16b, v22.16b\n"
    "umax v22.16b, v2.16b, v1.16b\n"
    "umax v18.16b, v27.16b, v21.16b\n"
    "umax v21.16b, v0.16b, v31.16b\n"
    "umax v17.16b, v26.16b, v20.16b\n"
    "umax v20.16b, v30.16b, v29.16b\n"
    "umax v16.16b, v25.16b, v24.16b\n"
    "umax v19.16b, v23.16b, v19.16b\n"
    "umax v18.16b, v22.16b, v18.16b\n"
    "umax v17.16b, v21.16b, v17.16b\n"
    "umax v16.16b, v20.16b, v16.16b\n"
    "umax v8.16b, v8.16b, v19.16b\n"
    "umax v7.16b, v7.16b, v18.16b\n"
    "umax v6.16b, v6.16b, v17.16b\n"
    "umax v5.16b, v5.16b, v16.16b\n"
    "4:"  // 4-vectors of channels: After loop
    "ands x21, %x[n_valid_cells], #0x3\n"
    "beq 6f\n"
    "5:"  // 4-vectors of channels: Single input loop
    "ldr x20, [x22], #0x8\n"
    "ldr q16, [x20, x27]\n"
    "subs x21, x21, #0x1\n"
    "umax v8.16b, v8.16b, v16.16b\n"
    "ldr q17, [x20, x26]\n"
    "ldr q16, [x20, x24]\n"
    "umax v7.16b, v7.16b, v17.16b\n"
    "umax v6.16b, v6.16b, v16.16b\n"
    "ldr q16, [x20, x23]\n"
    "umax v5.16b, v5.16b, v16.16b\n"
    "bgt 5b\n"
    "6:"  // 4-vectors of channels: Single input loop: End
    "add x20, %x[quant_params], %[offsetof_qp_input_offset]\n"
    "ld1r { v4.4s }, [x20]\n"
    "uxtl v23.8h, v8.8b\n"
    "uxtl2 v24.8h, v8.16b\n"
    "uxtl v22.8h, v7.8b\n"
    "uxtl2 v21.8h, v7.16b\n"
    "add x20, %x[quant_params], %[offsetof_qp_per_layer_left_shift]\n"
    "ld1r { v3.4s }, [x20]\n"
    "uxtl v20.8h, v6.8b\n"
    "uxtl2 v17.8h, v6.16b\n"
    "add x20, %x[quant_params], %[offsetof_qp_per_layer_mul]\n"
    "ld1r { v2.4s }, [x20]\n"
    "uxtl v19.8h, v5.8b\n"
    "uxtl2 v18.8h, v5.16b\n"
    "add x20, %x[quant_params], %[offsetof_qp_per_layer_right_shift]\n"
    "ld1r { v1.4s }, [x20]\n"
    "neg v4.4s, v4.4s\n"
    "saddw v0.4s, v4.4s, v23.4h\n"
    "add x20, %x[quant_params], %[offsetof_qp_output_offset]\n"
    "ld1r { v16.4s }, [x20]\n"
    "saddw2 v23.4s, v4.4s, v23.8h\n"
    "saddw v31.4s, v4.4s, v24.4h\n"
    "sub %x[n_channels], %x[n_channels], #0x40\n"
    "cmp %x[n_channels], #0x40\n"
    "saddw2 v30.4s, v4.4s, v24.8h\n"
    "saddw v29.4s, v4.4s, v22.4h\n"
    "saddw2 v22.4s, v4.4s, v22.8h\n"
    "saddw v28.4s, v4.4s, v21.4h\n"
    "saddw2 v21.4s, v4.4s, v21.8h\n"
    "saddw v27.4s, v4.4s, v20.4h\n"
    "saddw2 v20.4s, v4.4s, v20.8h\n"
    "saddw v26.4s, v4.4s, v17.4h\n"
    "saddw2 v17.4s, v4.4s, v17.8h\n"
    "saddw v25.4s, v4.4s, v19.4h\n"
    "saddw2 v19.4s, v4.4s, v19.8h\n"
    "saddw v24.4s, v4.4s, v18.4h\n"
    "saddw2 v18.4s, v4.4s, v18.8h\n"
    "srshl v0.4s, v0.4s, v3.4s\n"
    "srshl v23.4s, v23.4s, v3.4s\n"
    "srshl v31.4s, v31.4s, v3.4s\n"
    "srshl v30.4s, v30.4s, v3.4s\n"
    "srshl v29.4s, v29.4s, v3.4s\n"
    "srshl v22.4s, v22.4s, v3.4s\n"
    "srshl v28.4s, v28.4s, v3.4s\n"
    "srshl v21.4s, v21.4s, v3.4s\n"
    "srshl v27.4s, v27.4s, v3.4s\n"
    "srshl v20.4s, v20.4s, v3.4s\n"
    "srshl v26.4s, v26.4s, v3.4s\n"
    "srshl v17.4s, v17.4s, v3.4s\n"
    "srshl v25.4s, v25.4s, v3.4s\n"
    "srshl v19.4s, v19.4s, v3.4s\n"
    "srshl v24.4s, v24.4s, v3.4s\n"
    "srshl v18.4s, v18.4s, v3.4s\n"
    "sqrdmulh v0.4s, v0.4s, v2.4s\n"
    "sqrdmulh v23.4s, v23.4s, v2.4s\n"
    "sqrdmulh v31.4s, v31.4s, v2.4s\n"
    "sqrdmulh v30.4s, v30.4s, v2.4s\n"
    "sqrdmulh v29.4s, v29.4s, v2.4s\n"
    "sqrdmulh v22.4s, v22.4s, v2.4s\n"
    "sqrdmulh v28.4s, v28.4s, v2.4s\n"
    "sqrdmulh v21.4s, v21.4s, v2.4s\n"
    "sqrdmulh v27.4s, v27.4s, v2.4s\n"
    "sqrdmulh v20.4s, v20.4s, v2.4s\n"
    "sqrdmulh v26.4s, v26.4s, v2.4s\n"
    "sqrdmulh v17.4s, v17.4s, v2.4s\n"
    "sqrdmulh v25.4s, v25.4s, v2.4s\n"
    "sqrdmulh v19.4s, v19.4s, v2.4s\n"
    "sqrdmulh v24.4s, v24.4s, v2.4s\n"
    "sqrdmulh v18.4s, v18.4s, v2.4s\n"
    "srshl v0.4s, v0.4s, v1.4s\n"
    "srshl v23.4s, v23.4s, v1.4s\n"
    "srshl v31.4s, v31.4s, v1.4s\n"
    "srshl v30.4s, v30.4s, v1.4s\n"
    "srshl v29.4s, v29.4s, v1.4s\n"
    "srshl v22.4s, v22.4s, v1.4s\n"
    "srshl v28.4s, v28.4s, v1.4s\n"
    "srshl v21.4s, v21.4s, v1.4s\n"
    "srshl v27.4s, v27.4s, v1.4s\n"
    "srshl v20.4s, v20.4s, v1.4s\n"
    "srshl v26.4s, v26.4s, v1.4s\n"
    "srshl v17.4s, v17.4s, v1.4s\n"
    "srshl v25.4s, v25.4s, v1.4s\n"
    "srshl v19.4s, v19.4s, v1.4s\n"
    "srshl v24.4s, v24.4s, v1.4s\n"
    "srshl v18.4s, v18.4s, v1.4s\n"
    "add v0.4s, v0.4s, v16.4s\n"
    "add v23.4s, v23.4s, v16.4s\n"
    "add v31.4s, v31.4s, v16.4s\n"
    "add v30.4s, v30.4s, v16.4s\n"
    "add v29.4s, v29.4s, v16.4s\n"
    "add v22.4s, v22.4s, v16.4s\n"
    "add v28.4s, v28.4s, v16.4s\n"
    "add v21.4s, v21.4s, v16.4s\n"
    "add v27.4s, v27.4s, v16.4s\n"
    "add v20.4s, v20.4s, v16.4s\n"
    "add v26.4s, v26.4s, v16.4s\n"
    "add v17.4s, v17.4s, v16.4s\n"
    "add v25.4s, v25.4s, v16.4s\n"
    "add v19.4s, v19.4s, v16.4s\n"
    "add v24.4s, v24.4s, v16.4s\n"
    "add v18.4s, v18.4s, v16.4s\n"
    "movi v16.4s, #0x0\n"
    "smax v0.4s, v0.4s, v16.4s\n"
    "smax v23.4s, v23.4s, v16.4s\n"
    "smax v31.4s, v31.4s, v16.4s\n"
    "smax v30.4s, v30.4s, v16.4s\n"
    "smax v29.4s, v29.4s, v16.4s\n"
    "smax v22.4s, v22.4s, v16.4s\n"
    "smax v28.4s, v28.4s, v16.4s\n"
    "smax v21.4s, v21.4s, v16.4s\n"
    "smax v27.4s, v27.4s, v16.4s\n"
    "smax v20.4s, v20.4s, v16.4s\n"
    "smax v26.4s, v26.4s, v16.4s\n"
    "smax v17.4s, v17.4s, v16.4s\n"
    "smax v25.4s, v25.4s, v16.4s\n"
    "smax v19.4s, v19.4s, v16.4s\n"
    "smax v24.4s, v24.4s, v16.4s\n"
    "smax v18.4s, v18.4s, v16.4s\n"
    "movi v16.4s, #0xff\n"
    "smin v0.4s, v0.4s, v16.4s\n"
    "smin v23.4s, v23.4s, v16.4s\n"
    "smin v31.4s, v31.4s, v16.4s\n"
    "smin v30.4s, v30.4s, v16.4s\n"
    "smin v29.4s, v29.4s, v16.4s\n"
    "smin v22.4s, v22.4s, v16.4s\n"
    "smin v28.4s, v28.4s, v16.4s\n"
    "smin v21.4s, v21.4s, v16.4s\n"
    "smin v27.4s, v27.4s, v16.4s\n"
    "smin v20.4s, v20.4s, v16.4s\n"
    "smin v26.4s, v26.4s, v16.4s\n"
    "smin v17.4s, v17.4s, v16.4s\n"
    "smin v25.4s, v25.4s, v16.4s\n"
    "smin v19.4s, v19.4s, v16.4s\n"
    "smin v24.4s, v24.4s, v16.4s\n"
    "smin v18.4s, v18.4s, v16.4s\n"
    "uzp1 v23.16b, v0.16b, v23.16b\n"
    "uzp1 v16.16b, v31.16b, v30.16b\n"
    "uzp1 v22.16b, v29.16b, v22.16b\n"
    "uzp1 v21.16b, v28.16b, v21.16b\n"
    "uzp1 v20.16b, v27.16b, v20.16b\n"
    "uzp1 v17.16b, v26.16b, v17.16b\n"
    "uzp1 v19.16b, v25.16b, v19.16b\n"
    "uzp1 v18.16b, v24.16b, v18.16b\n"
    "uzp1 v16.16b, v23.16b, v16.16b\n"
    "str q16, [%x[outptr], x27]\n"
    "add x27, x27, #0x40\n"
    "uzp1 v16.16b, v22.16b, v21.16b\n"
    "uzp1 v17.16b, v20.16b, v17.16b\n"
    "str q16, [%x[outptr], x26]\n"
    "add x26, x26, #0x40\n"
    "uzp1 v16.16b, v19.16b, v18.16b\n"
    "str q17, [%x[outptr], x24]\n"
    "add x24, x24, #0x40\n"
    "str q16, [%x[outptr], x23]\n"
    "add x23, x23, #0x40\n"
    "bge 1b\n"
    "cbz %x[n_channels], 43f\n"
    "7:"  // Single vector of channels
    "cmp %x[n_channels], #0x10\n"
    "blt 14f\n"
    "8:"  // Single vector of channels: Loop
    "lsr x25, %x[n_valid_cells], #0x2\n"
    "movi v8.16b, #0x0\n"
    "mov x22, %x[inptrs]\n"
    "cbz x25, 11f\n"
    "ldp x21, x20, [x22, #0x0]\n"
    "ldr q4, [x21, x27]\n"
    "subs x25, x25, #0x1\n"
    "ldr q3, [x20, x27]\n"
    "ldp x21, x20, [x22, #0x10]\n"
    "add x22, x22, #0x20\n"
    "ldr q28, [x21, x27]\n"
    "ldr q22, [x20, x27]\n"
    "beq 10f\n"
    "9:"  // Single vector of channels: Loop: 4 inputs loop
    "umax v17.16b, v4.16b, v3.16b\n"
    "umax v16.16b, v28.16b, v22.16b\n"
    "ldp x21, x20, [x22, #0x0]\n"
    "ldr q4, [x21, x27]\n"
    "ldr q3, [x20, x27]\n"
    "umax v16.16b, v17.16b, v16.16b\n"
    "ldp x21, x20, [x22, #0x10]\n"
    "subs x25, x25, #0x1\n"
    "ldr q28, [x21, x27]\n"
    "ldr q22, [x20, x27]\n"
    "umax v8.16b, v8.16b, v16.16b\n"
    "add x22, x22, #0x20\n"
    "bgt 9b\n"
    "10:"  // Single vector of channels: Loop: 4 inputs tail
    "umax v17.16b, v4.16b, v3.16b\n"
    "umax v16.16b, v28.16b, v22.16b\n"
    "umax v16.16b, v17.16b, v16.16b\n"
    "umax v8.16b, v8.16b, v16.16b\n"
    "11:"  // Single vector of channels: Loop: After loop
    "ands x21, %x[n_valid_cells], #0x3\n"
    "beq 13f\n"
    "12:"  // Single vector of channels: Loop: Single input loop
    "ldr x20, [x22], #0x8\n"
    "ldr q16, [x20, x27]\n"
    "subs x21, x21, #0x1\n"
    "umax v8.16b, v8.16b, v16.16b\n"
    "bgt 12b\n"
    "13:"  // Single vector of channels: Loop: Single input loop: End
    "add x20, %x[quant_params], %[offsetof_qp_input_offset]\n"
    "ld1r { v18.4s }, [x20]\n"
    "uxtl v17.8h, v8.8b\n"
    "uxtl2 v16.8h, v8.16b\n"
    "neg v18.4s, v18.4s\n"
    "add x20, %x[quant_params], %[offsetof_qp_per_layer_left_shift]\n"
    "ld1r { v23.4s }, [x20]\n"
    "saddw v22.4s, v18.4s, v17.4h\n"
    "saddw2 v21.4s, v18.4s, v17.8h\n"
    "saddw v20.4s, v18.4s, v16.4h\n"
    "add x20, %x[quant_params], %[offsetof_qp_per_layer_mul]\n"
    "ld1r { v19.4s }, [x20]\n"
    "saddw2 v18.4s, v18.4s, v16.8h\n"
    "srshl v22.4s, v22.4s, v23.4s\n"
    "add x20, %x[quant_params], %[offsetof_qp_per_layer_right_shift]\n"
    "ld1r { v17.4s }, [x20]\n"
    "srshl v21.4s, v21.4s, v23.4s\n"
    "srshl v20.4s, v20.4s, v23.4s\n"
    "add x20, %x[quant_params], %[offsetof_qp_output_offset]\n"
    "ld1r { v16.4s }, [x20]\n"
    "srshl v18.4s, v18.4s, v23.4s\n"
    "sqrdmulh v22.4s, v22.4s, v19.4s\n"
    "sub %x[n_channels], %x[n_channels], #0x10\n"
    "cmp %x[n_channels], #0x10\n"
    "sqrdmulh v21.4s, v21.4s, v19.4s\n"
    "sqrdmulh v20.4s, v20.4s, v19.4s\n"
    "sqrdmulh v18.4s, v18.4s, v19.4s\n"
    "srshl v22.4s, v22.4s, v17.4s\n"
    "srshl v21.4s, v21.4s, v17.4s\n"
    "srshl v20.4s, v20.4s, v17.4s\n"
    "srshl v18.4s, v18.4s, v17.4s\n"
    "add v22.4s, v22.4s, v16.4s\n"
    "add v21.4s, v21.4s, v16.4s\n"
    "add v20.4s, v20.4s, v16.4s\n"
    "add v18.4s, v18.4s, v16.4s\n"
    "movi v16.4s, #0x0\n"
    "smax v22.4s, v22.4s, v16.4s\n"
    "smax v21.4s, v21.4s, v16.4s\n"
    "smax v20.4s, v20.4s, v16.4s\n"
    "smax v18.4s, v18.4s, v16.4s\n"
    "movi v16.4s, #0xff\n"
    "smin v22.4s, v22.4s, v16.4s\n"
    "smin v21.4s, v21.4s, v16.4s\n"
    "smin v20.4s, v20.4s, v16.4s\n"
    "smin v18.4s, v18.4s, v16.4s\n"
    "uzp1 v17.16b, v22.16b, v21.16b\n"
    "uzp1 v16.16b, v20.16b, v18.16b\n"
    "uzp1 v16.16b, v17.16b, v16.16b\n"
    "str q16, [%x[outptr], x27]\n"
    "add x27, x27, #0x10\n"
    "bge 8b\n"
    "cbz %x[n_channels], 43f\n"
    "14:"  // Oddments
    "lsr x25, %x[n_valid_cells], #0x2\n"
    "add %x[outptr], %x[outptr], x27\n"
    "movi v8.16b, #0x0\n"
    "mov x24, %x[inptrs]\n"
    "cbz x25, 24f\n"
    "15:"  // Oddments: 4 inputs loop
    "ldp x23, x22, [x24, #0x0]\n"
    "ldp x21, x20, [x24, #0x10]\n"
    "add x24, x24, #0x20\n"
    "add x23, x23, x27\n"
    "add x22, x22, x27\n"
    "add x21, x21, x27\n"
    "movi v4.16b, #0x0\n"
    "movi v3.16b, #0x0\n"
    "add x20, x20, x27\n"
    "movi v28.16b, #0x0\n"
    "movi v22.16b, #0x0\n"
    "tbz %x[n_channels], #3, 19f\n"
    "ldr d4, [x23], #0x8\n"
    "ldr d3, [x22], #0x8\n"
    "ldr d28, [x21], #0x8\n"
    "ldr d22, [x20], #0x8\n"
    "tbz %x[n_channels], #2, 17f\n"
    "ld1 { v4.s }[2], [x23], #0x4\n"
    "ld1 { v3.s }[2], [x22], #0x4\n"
    "ld1 { v28.s }[2], [x21], #0x4\n"
    "ld1 { v22.s }[2], [x20], #0x4\n"
    "tbz %x[n_channels], #1, 16f\n"
    "ld1 { v4.h }[6], [x23], #0x2\n"
    "ld1 { v3.h }[6], [x22], #0x2\n"
    "ld1 { v28.h }[6], [x21], #0x2\n"
    "ld1 { v22.h }[6], [x20], #0x2\n"
    "tbz %x[n_channels], #0, 23f\n"
    "ld1 { v4.b }[14], [x23], #0x1\n"
    "ld1 { v3.b }[14], [x22], #0x1\n"
    "ld1 { v28.b }[14], [x21], #0x1\n"
    "ld1 { v22.b }[14], [x20], #0x1\n"
    "b 23f\n"
    "16:"  // Oddments: 4 inputs loop: Load: Bit 3: Bit 2: Bit 1: Unset
    "tbz %x[n_channels], #0, 23f\n"
    "ld1 { v4.b }[12], [x23], #0x1\n"
    "ld1 { v3.b }[12], [x22], #0x1\n"
    "ld1 { v28.b }[12], [x21], #0x1\n"
    "ld1 { v22.b }[12], [x20], #0x1\n"
    "b 23f\n"
    "17:"  // Oddments: 4 inputs loop: Load: Bit 3: Bit 2: Unset
    "tbz %x[n_channels], #1, 18f\n"
    "ld1 { v4.h }[4], [x23], #0x2\n"
    "ld1 { v3.h }[4], [x22], #0x2\n"
    "ld1 { v28.h }[4], [x21], #0x2\n"
    "ld1 { v22.h }[4], [x20], #0x2\n"
    "tbz %x[n_channels], #0, 23f\n"
    "ld1 { v4.b }[10], [x23], #0x1\n"
    "ld1 { v3.b }[10], [x22], #0x1\n"
    "ld1 { v28.b }[10], [x21], #0x1\n"
    "ld1 { v22.b }[10], [x20], #0x1\n"
    "b 23f\n"
    "18:"  // Oddments: 4 inputs loop: Load: Bit 3: Bit 2: Unset: Bit 1: Unset
    "tbz %x[n_channels], #0, 23f\n"
    "ld1 { v4.b }[8], [x23], #0x1\n"
    "ld1 { v3.b }[8], [x22], #0x1\n"
    "ld1 { v28.b }[8], [x21], #0x1\n"
    "ld1 { v22.b }[8], [x20], #0x1\n"
    "b 23f\n"
    "19:"  // Oddments: 4 inputs loop: Load: Bit 3: Unset
    "tbz %x[n_channels], #2, 21f\n"
    "ldr s4, [x23], #0x4\n"
    "ldr s3, [x22], #0x4\n"
    "ldr s28, [x21], #0x4\n"
    "ldr s22, [x20], #0x4\n"
    "tbz %x[n_channels], #1, 20f\n"
    "ld1 { v4.h }[2], [x23], #0x2\n"
    "ld1 { v3.h }[2], [x22], #0x2\n"
    "ld1 { v28.h }[2], [x21], #0x2\n"
    "ld1 { v22.h }[2], [x20], #0x2\n"
    "tbz %x[n_channels], #0, 23f\n"
    "ld1 { v4.b }[6], [x23], #0x1\n"
    "ld1 { v3.b }[6], [x22], #0x1\n"
    "ld1 { v28.b }[6], [x21], #0x1\n"
    "ld1 { v22.b }[6], [x20], #0x1\n"
    "b 23f\n"
    "20:"  // Oddments: 4 inputs loop: Load: Bit 3: Unset: Bit 2: Bit 1: Unset
    "tbz %x[n_channels], #0, 23f\n"
    "ld1 { v4.b }[4], [x23], #0x1\n"
    "ld1 { v3.b }[4], [x22], #0x1\n"
    "ld1 { v28.b }[4], [x21], #0x1\n"
    "ld1 { v22.b }[4], [x20], #0x1\n"
    "b 23f\n"
    "21:"  // Oddments: 4 inputs loop: Load: Bit 3: Unset: Bit 2: Unset
    "tbz %x[n_channels], #1, 22f\n"
    "ldr h4, [x23], #0x2\n"
    "ldr h3, [x22], #0x2\n"
    "ldr h28, [x21], #0x2\n"
    "ldr h22, [x20], #0x2\n"
    "tbz %x[n_channels], #0, 23f\n"
    "ld1 { v4.b }[2], [x23], #0x1\n"
    "ld1 { v3.b }[2], [x22], #0x1\n"
    "ld1 { v28.b }[2], [x21], #0x1\n"
    "ld1 { v22.b }[2], [x20], #0x1\n"
    "b 23f\n"
    "22:"  // Oddments: 4 inputs loop: Load: Bit 3: Unset: Bit 2: Unset: Bit 1: Unset
    "tbz %x[n_channels], #0, 23f\n"
    "ldr b4, [x23], #0x1\n"
    "ldr b3, [x22], #0x1\n"
    "ldr b28, [x21], #0x1\n"
    "ldr b22, [x20], #0x1\n"
    "23:"  // Oddments: 4 inputs loop: Load: Bit 3: End
    "umax v17.16b, v4.16b, v3.16b\n"
    "umax v16.16b, v28.16b, v22.16b\n"
    "subs x25, x25, #0x1\n"
    "umax v16.16b, v17.16b, v16.16b\n"
    "umax v8.16b, v8.16b, v16.16b\n"
    "bgt 15b\n"
    "24:"  // Oddments: After loop
    "ands x21, %x[n_valid_cells], #0x3\n"
    "beq 34f\n"
    "25:"  // Oddments: Single input loop
    "ldr x23, [x24], #0x8\n"
    "add x23, x23, x27\n"
    "movi v4.16b, #0x0\n"
    "tbz %x[n_channels], #3, 29f\n"
    "ldr d4, [x23], #0x8\n"
    "tbz %x[n_channels], #2, 27f\n"
    "ld1 { v4.s }[2], [x23], #0x4\n"
    "tbz %x[n_channels], #1, 26f\n"
    "ld1 { v4.h }[6], [x23], #0x2\n"
    "tbz %x[n_channels], #0, 33f\n"
    "ld1 { v4.b }[14], [x23], #0x1\n"
    "b 33f\n"
    "26:"  // Oddments: Single input loop: Load: Bit 3: Bit 2: Bit 1: Unset
    "tbz %x[n_channels], #0, 33f\n"
    "ld1 { v4.b }[12], [x23], #0x1\n"
    "b 33f\n"
    "27:"  // Oddments: Single input loop: Load: Bit 3: Bit 2: Unset
    "tbz %x[n_channels], #1, 28f\n"
    "ld1 { v4.h }[4], [x23], #0x2\n"
    "tbz %x[n_channels], #0, 33f\n"
    "ld1 { v4.b }[10], [x23], #0x1\n"
    "b 33f\n"
    "28:"  // Oddments: Single input loop: Load: Bit 3: Bit 2: Unset: Bit 1: Unset
    "tbz %x[n_channels], #0, 33f\n"
    "ld1 { v4.b }[8], [x23], #0x1\n"
    "b 33f\n"
    "29:"  // Oddments: Single input loop: Load: Bit 3: Unset
    "tbz %x[n_channels], #2, 31f\n"
    "ldr s4, [x23], #0x4\n"
    "tbz %x[n_channels], #1, 30f\n"
    "ld1 { v4.h }[2], [x23], #0x2\n"
    "tbz %x[n_channels], #0, 33f\n"
    "ld1 { v4.b }[6], [x23], #0x1\n"
    "b 33f\n"
    "30:"  // Oddments: Single input loop: Load: Bit 3: Unset: Bit 2: Bit 1: Unset
    "tbz %x[n_channels], #0, 33f\n"
    "ld1 { v4.b }[4], [x23], #0x1\n"
    "b 33f\n"
    "31:"  // Oddments: Single input loop: Load: Bit 3: Unset: Bit 2: Unset
    "tbz %x[n_channels], #1, 32f\n"
    "ldr h4, [x23], #0x2\n"
    "tbz %x[n_channels], #0, 33f\n"
    "ld1 { v4.b }[2], [x23], #0x1\n"
    "b 33f\n"
    "32:"  // Oddments: Single input loop: Load: Bit 3: Unset: Bit 2: Unset: Bit 1: Unset
    "tbz %x[n_channels], #0, 33f\n"
    "ldr b4, [x23], #0x1\n"
    "33:"  // Oddments: Single input loop: Load: Bit 3: End
    "subs x21, x21, #0x1\n"
    "umax v8.16b, v8.16b, v4.16b\n"
    "bgt 25b\n"
    "34:"  // Oddments: Single input loop: End
    "add x20, %x[quant_params], %[offsetof_qp_input_offset]\n"
    "ld1r { v18.4s }, [x20]\n"
    "uxtl v17.8h, v8.8b\n"
    "uxtl2 v16.8h, v8.16b\n"
    "neg v18.4s, v18.4s\n"
    "add x20, %x[quant_params], %[offsetof_qp_per_layer_left_shift]\n"
    "ld1r { v23.4s }, [x20]\n"
    "saddw v22.4s, v18.4s, v17.4h\n"
    "saddw2 v21.4s, v18.4s, v17.8h\n"
    "saddw v20.4s, v18.4s, v16.4h\n"
    "add x20, %x[quant_params], %[offsetof_qp_per_layer_mul]\n"
    "ld1r { v19.4s }, [x20]\n"
    "saddw2 v18.4s, v18.4s, v16.8h\n"
    "srshl v22.4s, v22.4s, v23.4s\n"
    "add x20, %x[quant_params], %[offsetof_qp_per_layer_right_shift]\n"
    "ld1r { v17.4s }, [x20]\n"
    "srshl v21.4s, v21.4s, v23.4s\n"
    "srshl v20.4s, v20.4s, v23.4s\n"
    "add x20, %x[quant_params], %[offsetof_qp_output_offset]\n"
    "ld1r { v16.4s }, [x20]\n"
    "srshl v18.4s, v18.4s, v23.4s\n"
    "sqrdmulh v22.4s, v22.4s, v19.4s\n"
    "sqrdmulh v21.4s, v21.4s, v19.4s\n"
    "sqrdmulh v20.4s, v20.4s, v19.4s\n"
    "sqrdmulh v18.4s, v18.4s, v19.4s\n"
    "srshl v22.4s, v22.4s, v17.4s\n"
    "srshl v21.4s, v21.4s, v17.4s\n"
    "srshl v20.4s, v20.4s, v17.4s\n"
    "srshl v18.4s, v18.4s, v17.4s\n"
    "add v22.4s, v22.4s, v16.4s\n"
    "add v21.4s, v21.4s, v16.4s\n"
    "add v20.4s, v20.4s, v16.4s\n"
    "add v18.4s, v18.4s, v16.4s\n"
    "movi v16.4s, #0x0\n"
    "smax v22.4s, v22.4s, v16.4s\n"
    "smax v21.4s, v21.4s, v16.4s\n"
    "smax v20.4s, v20.4s, v16.4s\n"
    "smax v18.4s, v18.4s, v16.4s\n"
    "movi v16.4s, #0xff\n"
    "smin v22.4s, v22.4s, v16.4s\n"
    "smin v21.4s, v21.4s, v16.4s\n"
    "smin v20.4s, v20.4s, v16.4s\n"
    "smin v18.4s, v18.4s, v16.4s\n"
    "uzp1 v17.16b, v22.16b, v21.16b\n"
    "uzp1 v16.16b, v20.16b, v18.16b\n"
    "uzp1 v16.16b, v17.16b, v16.16b\n"
    "tbz %x[n_channels], #3, 38f\n"
    "st1 { v16.d }[0], [%x[outptr]], #0x8\n"
    "tbz %x[n_channels], #2, 36f\n"
    "st1 { v16.s }[2], [%x[outptr]], #0x4\n"
    "tbz %x[n_channels], #1, 35f\n"
    "st1 { v16.h }[6], [%x[outptr]], #0x2\n"
    "tbz %x[n_channels], #0, 42f\n"
    "st1 { v16.b }[14], [%x[outptr]], #0x1\n"
    "b 42f\n"
    "35:"  // Oddments: Store: Bit 3: Bit 2: Bit 1: Unset
    "tbz %x[n_channels], #0, 42f\n"
    "st1 { v16.b }[12], [%x[outptr]], #0x1\n"
    "b 42f\n"
    "36:"  // Oddments: Store: Bit 3: Bit 2: Unset
    "tbz %x[n_channels], #1, 37f\n"
    "st1 { v16.h }[4], [%x[outptr]], #0x2\n"
    "tbz %x[n_channels], #0, 42f\n"
    "st1 { v16.b }[10], [%x[outptr]], #0x1\n"
    "b 42f\n"
    "37:"  // Oddments: Store: Bit 3: Bit 2: Unset: Bit 1: Unset
    "tbz %x[n_channels], #0, 42f\n"
    "st1 { v16.b }[8], [%x[outptr]], #0x1\n"
    "b 42f\n"
    "38:"  // Oddments: Store: Bit 3: Unset
    "tbz %x[n_channels], #2, 40f\n"
    "st1 { v16.s }[0], [%x[outptr]], #0x4\n"
    "tbz %x[n_channels], #1, 39f\n"
    "st1 { v16.h }[2], [%x[outptr]], #0x2\n"
    "tbz %x[n_channels], #0, 42f\n"
    "st1 { v16.b }[6], [%x[outptr]], #0x1\n"
    "b 42f\n"
    "39:"  // Oddments: Store: Bit 3: Unset: Bit 2: Bit 1: Unset
    "tbz %x[n_channels], #0, 42f\n"
    "st1 { v16.b }[4], [%x[outptr]], #0x1\n"
    "b 42f\n"
    "40:"  // Oddments: Store: Bit 3: Unset: Bit 2: Unset
    "tbz %x[n_channels], #1, 41f\n"
    "st1 { v16.h }[0], [%x[outptr]], #0x2\n"
    "tbz %x[n_channels], #0, 42f\n"
    "st1 { v16.b }[2], [%x[outptr]], #0x1\n"
    "b 42f\n"
    "41:"  // Oddments: Store: Bit 3: Unset: Bit 2: Unset: Bit 1: Unset
    "tbz %x[n_channels], #0, 42f\n"
    "st1 { v16.b }[0], [%x[outptr]], #0x1\n"
    "42:"  // Oddments: Store: Bit 3: End
    "43:"  // End
    : [n_channels] "+&r" (n_channels), [outptr] "+&r" (outptr)
    : [inptrs] "r" (inptrs), [n_valid_cells] "r" (n_valid_cells), [offsetof_qp_input_offset] "I" (offsetof(Requantize32, input_offset)), [offsetof_qp_output_offset] "I" (offsetof(Requantize32, output_offset)), [offsetof_qp_per_layer_left_shift] "I" (offsetof(Requantize32, per_layer_left_shift)), [offsetof_qp_per_layer_mul] "I" (offsetof(Requantize32, per_layer_mul)), [offsetof_qp_per_layer_right_shift] "I" (offsetof(Requantize32, per_layer_right_shift)), [quant_params] "r" (&qp)
    : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
  );
}

}  // namespace pooling
}  // namespace arm_conv

#endif  // defined(__aarch64__)