src/operators/max-pooling-nhwc.c

// Copyright (c) Facebook, Inc. and its affiliates.
// All rights reserved.
//
// Copyright 2019 Google LLC
//
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree.

#include <assert.h>
#include <math.h>
#include <stdbool.h>
#include <stddef.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>

#include <fp16.h>

#include <xnnpack.h>
#include <xnnpack/allocator.h>
#include <xnnpack/common.h>
#include <xnnpack/indirection.h>
#include <xnnpack/log.h>
#include <xnnpack/math.h>
#include <xnnpack/operator.h>
#include <xnnpack/microparams-init.h>
#include <xnnpack/params.h>


static inline size_t compute_output_dimension_with_tf_same_padding(
    size_t input_dimension,
    size_t stride_dimension)
{
  return divide_round_up(input_dimension, stride_dimension);
}

static enum xnn_status create_max_pooling2d_nhwc(
    uint32_t input_padding_top,
    uint32_t input_padding_right,
    uint32_t input_padding_bottom,
    uint32_t input_padding_left,
    uint32_t pooling_height,
    uint32_t pooling_width,
    uint32_t stride_height,
    uint32_t stride_width,
    uint32_t dilation_height,
    uint32_t dilation_width,
    size_t channels,
    size_t input_pixel_stride,
    size_t output_pixel_stride,
    uint32_t flags,
    const void* params,
    size_t params_size,
    uint32_t datatype_init_flags,
    enum xnn_operator_type operator_type,
    xnn_operator_t* max_pooling_op_out)
{
  xnn_operator_t max_pooling_op = NULL;
  enum xnn_status status = xnn_status_uninitialized;

  if ((xnn_params.init_flags & XNN_INIT_FLAG_XNNPACK) == 0) {
    xnn_log_error("failed to setup %s operator: XNNPACK is not initialized",
      xnn_operator_type_to_string(operator_type));
    return xnn_status_uninitialized;
  }

  status = xnn_status_unsupported_hardware;

  if ((xnn_params.init_flags & datatype_init_flags) != datatype_init_flags) {
    xnn_log_error(
      "failed to create %s operator: operations on data type are not supported",
      xnn_operator_type_to_string(operator_type));
    goto error;
  }

  status = xnn_status_invalid_parameter;

  const uint32_t pooling_size = pooling_height * pooling_width;
  if (pooling_size == 0) {
    xnn_log_error(
      "failed to create %s operator with %" PRIu32 "x%" PRIu32 " pooling size: "
      "pooling size dimensions must be non-zero",
      xnn_operator_type_to_string(operator_type),
      pooling_width, pooling_height);
    goto error;
  }

  if (pooling_size == 1) {
    xnn_log_error(
      "failed to create %s operator with 1 pooling element: 1x1 pooling is meaningless",
      xnn_operator_type_to_string(operator_type));
    goto error;
  }

  if (stride_height == 0 || stride_width == 0) {
    xnn_log_error(
      "failed to create %s operator with %" PRIu32 "x%" PRIu32 " stride: stride dimensions must be non-zero",
      xnn_operator_type_to_string(operator_type), stride_width, stride_height);
    goto error;
  }

  if (dilation_height == 0 || dilation_width == 0) {
    xnn_log_error(
      "failed to create %s operator with %" PRIu32 "x%" PRIu32 " dilation: dilation dimensions must be non-zero",
      xnn_operator_type_to_string(operator_type), dilation_width, dilation_height);
    goto error;
  }

  if (stride_height > pooling_height) {
    xnn_log_error(
      "failed to define %s operator with %" PRIu32 " stride height: must be less than pooling height %" PRIu32,
      xnn_operator_type_to_string(operator_type), stride_height, pooling_height);
    return xnn_status_invalid_parameter;
  }

  if (stride_width > pooling_width) {
    xnn_log_error(
      "failed to define %s operator with %" PRIu32 " stride width: must be less than pooling width %" PRIu32,
      xnn_operator_type_to_string(operator_type), stride_width, pooling_width);
    return xnn_status_invalid_parameter;
  }

  if (channels == 0) {
    xnn_log_error(
      "failed to create %s operator with %zu channels: number of channels must be non-zero",
      xnn_operator_type_to_string(operator_type), channels);
    goto error;
  }

  if (input_pixel_stride < channels) {
    xnn_log_error(
      "failed to create %s operator with input pixel stride of %zu: "
      "stride must be at least as large as the number of channels (%zu)",
      xnn_operator_type_to_string(operator_type), input_pixel_stride, channels);
    goto error;
  }

  if (output_pixel_stride < channels) {
    xnn_log_error(
      "failed to create %s operator with output pixel stride of %zu: "
      "stride must be at least as large as the number of channels (%zu)",
      xnn_operator_type_to_string(operator_type), output_pixel_stride, channels);
    goto error;
  }

  const bool any_padding = (input_padding_left | input_padding_top | input_padding_right | input_padding_bottom) != 0;
  if ((flags & XNN_FLAG_TENSORFLOW_SAME_PADDING) != 0) {
    if (any_padding) {
      xnn_log_error(
        "failed to create %s operator with %" PRIu32 "+%" PRIu32 "x%" PRIu32 "+%" PRIu32" padding: "
        "TensorFlow SAME padding can't be combined with explicit padding specification",
        xnn_operator_type_to_string(operator_type),
        input_padding_top, input_padding_left, input_padding_bottom, input_padding_right);
      goto error;
    }
  }

  status = xnn_status_out_of_memory;

  max_pooling_op = xnn_allocate_zero_simd_memory(sizeof(struct xnn_operator));
  if (max_pooling_op == NULL) {
    xnn_log_error(
      "failed to allocate %zu bytes for %s operator descriptor",
      sizeof(struct xnn_operator), xnn_operator_type_to_string(operator_type));
    goto error;
  }

  max_pooling_op->padding_top = input_padding_top;
  max_pooling_op->padding_right = input_padding_right;
  max_pooling_op->padding_bottom = input_padding_bottom;
  max_pooling_op->padding_left = input_padding_left;

  max_pooling_op->kernel_height = pooling_height;
  max_pooling_op->kernel_width = pooling_width;
  max_pooling_op->stride_height = stride_height;
  max_pooling_op->stride_width = stride_width;
  max_pooling_op->dilation_height = dilation_height;
  max_pooling_op->dilation_width = dilation_width;
  max_pooling_op->channels = channels;
  max_pooling_op->input_pixel_stride = input_pixel_stride;
  max_pooling_op->output_pixel_stride = output_pixel_stride;

  memcpy(&max_pooling_op->params, params, params_size);
  max_pooling_op->type = operator_type;
  max_pooling_op->flags = flags;

  max_pooling_op->state = xnn_run_state_invalid;

  *max_pooling_op_out = max_pooling_op;
  return xnn_status_success;

error:
  xnn_delete_operator(max_pooling_op);
  return status;
}

static enum xnn_status setup_max_pooling2d_nhwc(
  xnn_operator_t max_pooling_op,
  enum xnn_operator_type expected_operator_type,
  size_t batch_size,
  size_t input_height,
  size_t input_width,
  const void* input,
  void* output,
  uint32_t log2_input_element_size,
  uint32_t log2_output_element_size,
  struct maxpool_parameters maxpool[restrict XNN_MIN_ELEMENTS(1)],
  const void* params,
  size_t params_size,
  size_t num_threads)
{
  if (max_pooling_op->type != expected_operator_type) {
    xnn_log_error("failed to setup operator: operator type mismatch (expected %s, got %s)",
      xnn_operator_type_to_string(expected_operator_type),
      xnn_operator_type_to_string(max_pooling_op->type));
    return xnn_status_invalid_parameter;
  }
  max_pooling_op->state = xnn_run_state_invalid;

  if ((xnn_params.init_flags & XNN_INIT_FLAG_XNNPACK) == 0) {
    xnn_log_error(
      "failed to setup %s operator: XNNPACK is not initialized",
      xnn_operator_type_to_string(max_pooling_op->type));
    return xnn_status_uninitialized;
  }

  if (input_width == 0 || input_height == 0) {
    xnn_log_error(
      "failed to setup %s operator with %zux%zu input: input dimensions must be non-zero",
      xnn_operator_type_to_string(max_pooling_op->type), input_width, input_height);
    return xnn_status_invalid_parameter;
  }

  if (batch_size == 0) {
    max_pooling_op->state = xnn_run_state_skip;
    return xnn_status_success;
  }

  max_pooling_op->input_height = input_height;
  max_pooling_op->input_width = input_width;
  max_pooling_op->input = input;

  if (max_pooling_op->flags & XNN_FLAG_TENSORFLOW_SAME_PADDING) {
    max_pooling_op->output_height = compute_output_dimension_with_tf_same_padding(
        input_height, max_pooling_op->stride_height);
    max_pooling_op->output_width = compute_output_dimension_with_tf_same_padding(
        input_width, max_pooling_op->stride_width);

    const uint32_t effective_kernel_height = (max_pooling_op->kernel_height - 1) * max_pooling_op->dilation_height + 1;
    const uint32_t effective_kernel_width = (max_pooling_op->kernel_width - 1) * max_pooling_op->dilation_width + 1;
    const uint32_t total_padding_height =
      doz((max_pooling_op->output_height - 1) * max_pooling_op->stride_height + effective_kernel_height, input_height);
    const uint32_t total_padding_width =
      doz((max_pooling_op->output_width - 1) * max_pooling_op->stride_width + effective_kernel_width, input_width);
    max_pooling_op->padding_top = total_padding_height / 2;
    max_pooling_op->padding_left = total_padding_width / 2;
    max_pooling_op->padding_bottom = total_padding_height - max_pooling_op->padding_top;
    max_pooling_op->padding_right = total_padding_width - max_pooling_op->padding_left;
  } else {
    max_pooling_op->output_height = xnn_compute_convolution_output_dimension(
        max_pooling_op->padding_top + input_height + max_pooling_op->padding_bottom,
        max_pooling_op->kernel_height,
        max_pooling_op->dilation_height,
        max_pooling_op->stride_height);
    max_pooling_op->output_width = xnn_compute_convolution_output_dimension(
        max_pooling_op->padding_left + input_width + max_pooling_op->padding_right,
        max_pooling_op->kernel_width,
        max_pooling_op->dilation_width,
        max_pooling_op->stride_width);
  }

  const size_t pooling_height = max_pooling_op->kernel_height;
  const size_t pooling_width = max_pooling_op->kernel_width;
  const size_t pooling_size = pooling_height * pooling_width;
  const size_t output_height = max_pooling_op->output_height;
  const size_t output_width = max_pooling_op->output_width;
  const uint32_t mr = maxpool->mr;

  const size_t step_width =
    max_pooling_op->dilation_width > 1 ? pooling_width : min(max_pooling_op->stride_width, pooling_width);
  const size_t step_height = pooling_size + (output_width - 1) * step_width * pooling_height;

  if (input_height != max_pooling_op->last_input_height ||
      input_width != max_pooling_op->last_input_width)
  {
    // Micro-kernel may read up to (mr - 1) elements after the end of indirection buffer.
    const size_t indirection_buffer_size = sizeof(void*) * ((mr - 1) + output_height * step_height);
    const void** indirection_buffer =
      (const void**) xnn_reallocate_memory(max_pooling_op->indirection_buffer, indirection_buffer_size);
    if (indirection_buffer == NULL) {
      xnn_log_error("failed to allocate %zu bytes for indirection buffer", indirection_buffer_size);
      return xnn_status_out_of_memory;
    }
    max_pooling_op->indirection_buffer = indirection_buffer;

    xnn_indirection_init_maxpool2d(max_pooling_op, step_height, step_width, log2_input_element_size);

    max_pooling_op->last_input = input;
    max_pooling_op->last_input_height = input_height;
    max_pooling_op->last_input_width = input_width;
  }

  const uint32_t qr = maxpool->qr;
  const size_t channels = max_pooling_op->channels;

  const size_t indirect_input_height_stride = step_height * sizeof(void*);
  const size_t output_width_stride = max_pooling_op->output_pixel_stride << log2_output_element_size;
  const size_t output_height_stride = output_width * output_width_stride;
  const size_t multipass_adjustment = round_up(doz(pooling_size, mr), qr) + mr;

  max_pooling_op->context.max_pooling = (struct max_pooling_context) {
    .indirect_input = max_pooling_op->indirection_buffer,
    .indirect_input_height_stride = indirect_input_height_stride,
    .input_offset = (size_t) ((uintptr_t) input - (uintptr_t) max_pooling_op->last_input),
    .input_batch_stride = (input_height * input_width * max_pooling_op->input_pixel_stride) << log2_input_element_size,
    .output = output,
    .output_batch_stride = output_height * output_height_stride,
    .output_height_stride = output_height_stride,
    .output_width = output_width,
    .pooling_size = pooling_size,
    .channels = channels,
    .input_increment = (pooling_height * step_width - multipass_adjustment) * sizeof(void*),
    .output_increment = output_width_stride - (channels << log2_output_element_size),
    .ukernel = maxpool->ukernel,
  };
  memcpy(&max_pooling_op->context.max_pooling.params, params, params_size);

  max_pooling_op->compute.type = xnn_parallelization_type_2d;
  max_pooling_op->compute.task_2d = (pthreadpool_task_2d_t) xnn_compute_max_pooling;
  max_pooling_op->compute.range[0] = batch_size;
  max_pooling_op->compute.range[1] = output_height;
  max_pooling_op->state = xnn_run_state_ready;

  return xnn_status_success;
}

enum xnn_status xnn_create_max_pooling2d_nhwc_s8(
    uint32_t input_padding_top,
    uint32_t input_padding_right,
    uint32_t input_padding_bottom,
    uint32_t input_padding_left,
    uint32_t pooling_height,
    uint32_t pooling_width,
    uint32_t stride_height,
    uint32_t stride_width,
    uint32_t dilation_height,
    uint32_t dilation_width,
    size_t channels,
    size_t input_pixel_stride,
    size_t output_pixel_stride,
    int8_t output_min,
    int8_t output_max,
    uint32_t flags,
    xnn_operator_t* max_pooling_op_out)
{
  if (output_min >= output_max) {
    xnn_log_error(
      "failed to create %s operator with [%" PRId8 ", %" PRId8 "] output range: range min must be below range max",
      xnn_operator_type_to_string(xnn_operator_type_max_pooling_nhwc_s8), output_min, output_max);
    return xnn_status_invalid_parameter;
  }

  union xnn_s8_minmax_params params;
  xnn_params.s8.maxpool.init.s8(&params, output_min, output_max);
  return create_max_pooling2d_nhwc(
    input_padding_top, input_padding_right, input_padding_bottom, input_padding_left,
    pooling_height, pooling_width,
    stride_height, stride_width,
    dilation_height, dilation_width,
    channels, input_pixel_stride, output_pixel_stride,
    flags,
    &params, sizeof(params), XNN_INIT_FLAG_S8,
    xnn_operator_type_max_pooling_nhwc_s8,
    max_pooling_op_out);
}

enum xnn_status xnn_create_max_pooling2d_nhwc_u8(
    uint32_t input_padding_top,
    uint32_t input_padding_right,
    uint32_t input_padding_bottom,
    uint32_t input_padding_left,
    uint32_t pooling_height,
    uint32_t pooling_width,
    uint32_t stride_height,
    uint32_t stride_width,
    uint32_t dilation_height,
    uint32_t dilation_width,
    size_t channels,
    size_t input_pixel_stride,
    size_t output_pixel_stride,
    uint8_t output_min,
    uint8_t output_max,
    uint32_t flags,
    xnn_operator_t* max_pooling_op_out)
{
  if (output_min >= output_max) {
    xnn_log_error(
      "failed to create %s operator with [%" PRIu8 ", %" PRIu8 "] output range: range min must be below range max",
      xnn_operator_type_to_string(xnn_operator_type_max_pooling_nhwc_u8), output_min, output_max);
    return xnn_status_invalid_parameter;
  }

  union xnn_u8_minmax_params params;
  xnn_params.u8.maxpool.init.u8(&params, output_min, output_max);
  return create_max_pooling2d_nhwc(
    input_padding_top, input_padding_right, input_padding_bottom, input_padding_left,
    pooling_height, pooling_width,
    stride_height, stride_width,
    dilation_height, dilation_width,
    channels, input_pixel_stride, output_pixel_stride,
    flags,
    &params, sizeof(params), XNN_INIT_FLAG_U8,
    xnn_operator_type_max_pooling_nhwc_u8,
    max_pooling_op_out);
}

enum xnn_status xnn_create_max_pooling2d_nhwc_f32(
    uint32_t input_padding_top,
    uint32_t input_padding_right,
    uint32_t input_padding_bottom,
    uint32_t input_padding_left,
    uint32_t pooling_height,
    uint32_t pooling_width,
    uint32_t stride_height,
    uint32_t stride_width,
    uint32_t dilation_height,
    uint32_t dilation_width,
    size_t channels,
    size_t input_pixel_stride,
    size_t output_pixel_stride,
    float output_min,
    float output_max,
    uint32_t flags,
    xnn_operator_t* max_pooling_op_out)
{
  if (isnan(output_min)) {
    xnn_log_error(
      "failed to create %s with NaN output lower bound: lower bound must be non-NaN",
      xnn_operator_type_to_string(xnn_operator_type_max_pooling_nhwc_f32));
    return xnn_status_invalid_parameter;
  }

  if (isnan(output_max)) {
    xnn_log_error(
      "failed to create %s with NaN output upper bound: upper bound must be non-NaN",
      xnn_operator_type_to_string(xnn_operator_type_max_pooling_nhwc_f32));
    return xnn_status_invalid_parameter;
  }

  if (output_min >= output_max) {
    xnn_log_error(
      "failed to create %s with [%.7g, %.7g] output range: lower bound must be below upper bound",
      xnn_operator_type_to_string(xnn_operator_type_max_pooling_nhwc_f32), output_min, output_max);
    return xnn_status_invalid_parameter;
  }

  union xnn_f32_minmax_params params;
  xnn_params.f32.maxpool.init.f32(&params, output_min, output_max);
  return create_max_pooling2d_nhwc(
    input_padding_top, input_padding_right, input_padding_bottom, input_padding_left,
    pooling_height, pooling_width,
    stride_height, stride_width,
    dilation_height, dilation_width,
    channels, input_pixel_stride, output_pixel_stride,
    flags,
    &params, sizeof(params), XNN_INIT_FLAG_F32,
    xnn_operator_type_max_pooling_nhwc_f32,
    max_pooling_op_out);
}

enum xnn_status xnn_create_max_pooling2d_nhwc_f16(
    uint32_t input_padding_top,
    uint32_t input_padding_right,
    uint32_t input_padding_bottom,
    uint32_t input_padding_left,
    uint32_t pooling_height,
    uint32_t pooling_width,
    uint32_t stride_height,
    uint32_t stride_width,
    uint32_t dilation_height,
    uint32_t dilation_width,
    size_t channels,
    size_t input_pixel_stride,
    size_t output_pixel_stride,
    float output_min,
    float output_max,
    uint32_t flags,
    xnn_operator_t* max_pooling_op_out)
{
  if (isnan(output_min)) {
    xnn_log_error(
      "failed to create %s with NaN output lower bound: lower bound must be non-NaN",
      xnn_operator_type_to_string(xnn_operator_type_max_pooling_nhwc_f16));
    return xnn_status_invalid_parameter;
  }

  if (isnan(output_max)) {
    xnn_log_error(
      "failed to create %s with NaN output upper bound: upper bound must be non-NaN",
      xnn_operator_type_to_string(xnn_operator_type_max_pooling_nhwc_f16));
    return xnn_status_invalid_parameter;
  }

  const uint16_t output_min_as_half = fp16_ieee_from_fp32_value(output_min);
  const uint16_t output_max_as_half = fp16_ieee_from_fp32_value(output_max);
  output_min = fp16_ieee_to_fp32_value(output_min_as_half);
  output_max = fp16_ieee_to_fp32_value(output_max_as_half);
  if (output_min >= output_max) {
    xnn_log_error(
      "failed to create %s operator with [%.7g, %.7g] output range: lower bound must be below upper bound",
      xnn_operator_type_to_string(xnn_operator_type_max_pooling_nhwc_f16), output_min, output_max);
    return xnn_status_invalid_parameter;
  }

  union xnn_f16_minmax_params params;
  if (xnn_params.f16.maxpool.init.f16 != NULL) {
    xnn_params.f16.maxpool.init.f16(&params, output_min_as_half, output_max_as_half);
  }
  return create_max_pooling2d_nhwc(
    input_padding_top, input_padding_right, input_padding_bottom, input_padding_left,
    pooling_height, pooling_width,
    stride_height, stride_width,
    dilation_height, dilation_width,
    channels, input_pixel_stride, output_pixel_stride,
    flags,
    &params, sizeof(params), XNN_INIT_FLAG_F16,
    xnn_operator_type_max_pooling_nhwc_f16,
    max_pooling_op_out);
}

enum xnn_status xnn_setup_max_pooling2d_nhwc_s8(
    xnn_operator_t max_pooling_op,
    size_t batch_size,
    size_t input_height,
    size_t input_width,
    const int8_t* input,
    int8_t* output,
    pthreadpool_t threadpool)
{
  return setup_max_pooling2d_nhwc(
    max_pooling_op, xnn_operator_type_max_pooling_nhwc_s8,
    batch_size, input_height, input_width,
    input, output,
    0 /* log2(sizeof(input element)) = log2(sizeof(int8_t)) */,
    0 /* log2(sizeof(output element)) = log2(sizeof(int8_t)) */,
    &xnn_params.s8.maxpool,
    &max_pooling_op->params.s8_minmax, sizeof(max_pooling_op->params.s8_minmax),
    pthreadpool_get_threads_count(threadpool));
}

enum xnn_status xnn_setup_max_pooling2d_nhwc_u8(
    xnn_operator_t max_pooling_op,
    size_t batch_size,
    size_t input_height,
    size_t input_width,
    const uint8_t* input,
    uint8_t* output,
    pthreadpool_t threadpool)
{
  return setup_max_pooling2d_nhwc(
    max_pooling_op, xnn_operator_type_max_pooling_nhwc_u8,
    batch_size, input_height, input_width,
    input, output,
    0 /* log2(sizeof(input element)) = log2(sizeof(uint8_t)) */,
    0 /* log2(sizeof(output element)) = log2(sizeof(uint8_t)) */,
    &xnn_params.u8.maxpool,
    &max_pooling_op->params.u8_minmax, sizeof(max_pooling_op->params.u8_minmax),
    pthreadpool_get_threads_count(threadpool));
}

enum xnn_status xnn_setup_max_pooling2d_nhwc_f16(
    xnn_operator_t max_pooling_op,
    size_t batch_size,
    size_t input_height,
    size_t input_width,
    const void* input,
    void* output,
    pthreadpool_t threadpool)
{
  return setup_max_pooling2d_nhwc(
    max_pooling_op, xnn_operator_type_max_pooling_nhwc_f16,
    batch_size, input_height, input_width,
    input, output,
    1 /* log2(sizeof(input element)) = log2(sizeof(uint16_t)) */,
    1 /* log2(sizeof(output element)) = log2(sizeof(uint16_t)) */,
    &xnn_params.f16.maxpool,
    &max_pooling_op->params.f16_minmax, sizeof(max_pooling_op->params.f16_minmax),
    pthreadpool_get_threads_count(threadpool));
}

enum xnn_status xnn_setup_max_pooling2d_nhwc_f32(
    xnn_operator_t max_pooling_op,
    size_t batch_size,
    size_t input_height,
    size_t input_width,
    const float* input,
    float* output,
    pthreadpool_t threadpool)
{
  return setup_max_pooling2d_nhwc(
    max_pooling_op, xnn_operator_type_max_pooling_nhwc_f32,
    batch_size, input_height, input_width,
    input, output,
    2 /* log2(sizeof(input element)) = log2(sizeof(float)) */,
    2 /* log2(sizeof(output element)) = log2(sizeof(float)) */,
    &xnn_params.f32.maxpool,
    &max_pooling_op->params.f32_minmax, sizeof(max_pooling_op->params.f32_minmax),
    pthreadpool_get_threads_count(threadpool));
}