// SPDX-License-Identifier: AGPL-3.0-or-later
// Permute - Reorder tensor dimensions (f64 canonical)
// Transposes tensor according to dimension permutation
//
// Example: permute([B, C, H, W], [0, 2, 3, 1]) → [B, H, W, C]
//
// Algorithm:
// For each output index, compute corresponding input index using permutation
struct Params {
total_size: u32,
num_dims: u32,
_pad1: u32,
_pad2: u32,
}
@group(0) @binding(0) var<uniform> params: Params;
@group(0) @binding(1) var<storage, read> input: array<f64>;
@group(0) @binding(2) var<storage, read> input_shape: array<u32>; // Original shape
@group(0) @binding(3) var<storage, read> output_shape: array<u32>; // Permuted shape
@group(0) @binding(4) var<storage, read> permutation: array<u32>; // Dimension mapping
@group(0) @binding(5) var<storage, read> input_strides: array<u32>; // Strides for input
@group(0) @binding(6) var<storage, read_write> output: array<f64>;
@compute @workgroup_size(256)
fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
let out_idx = global_id.x;
if (out_idx >= params.total_size) {
return;
}
// Compute multi-dimensional output indices
var temp_idx = out_idx;
var in_idx = 0u;
for (var i = 0u; i < params.num_dims; i++) {
let dim_idx = temp_idx / output_shape[i + 1u]; // Division by stride
temp_idx = temp_idx % output_shape[i + 1u];
// Map to input dimension using permutation
let in_dim = permutation[i];
in_idx += dim_idx * input_strides[in_dim];
}
output[out_idx] = input[in_idx];
}