// SPDX-License-Identifier: AGPL-3.0-or-later
// Adaptive Max Pooling 2D
// Pools input to a specific output size using maximum operation
//
// Automatically computes pooling regions to produce desired output size.
//
// Input shape: [batch, channels, in_height, in_width]
// Output shape: [batch, channels, out_height, out_width]
//
// Used in: Classification, SPPNet, flexible architectures
// Benefits: Variable input sizes, captures most salient features
@group(0) @binding(0) var<storage, read> input: array<f64>;
@group(0) @binding(1) var<storage, read_write> output: array<f64>;
@group(0) @binding(2) var<uniform> params: Params;
struct Params {
batch: u32,
channels: u32,
in_height: u32,
in_width: u32,
out_height: u32,
out_width: u32,
}
@compute @workgroup_size(16, 16)
fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
let out_x = global_id.x;
let out_y = global_id.y;
let batch_channel = global_id.z;
if out_x >= params.out_width || out_y >= params.out_height {
return;
}
if batch_channel >= params.batch * params.channels {
return;
}
let b = batch_channel / params.channels;
let c = batch_channel % params.channels;
// Compute adaptive pooling window
let start_h = (out_y * params.in_height) / params.out_height;
let end_h = ((out_y + 1u) * params.in_height) / params.out_height;
let start_w = (out_x * params.in_width) / params.out_width;
let end_w = ((out_x + 1u) * params.in_width) / params.out_width;
// Max over the adaptive window
var max_val: f64 = -1e308; // Very large negative number
for (var h = start_h; h < end_h; h = h + 1u) {
for (var w = start_w; w < end_w; w = w + 1u) {
let input_idx = b * (params.channels * params.in_height * params.in_width) +
c * (params.in_height * params.in_width) +
h * params.in_width +
w;
max_val = max(max_val, input[input_idx]);
}
}
let output_idx = b * (params.channels * params.out_height * params.out_width) +
c * (params.out_height * params.out_width) +
out_y * params.out_width +
out_x;
output[output_idx] = max_val;
}