// SPDX-License-Identifier: AGPL-3.0-or-later
// Argmax Reduction - Find index of maximum value over all elements (f64 canonical)
// CUDA equivalent: thrust::reduce with argmax operation
// Algorithm: Tree reduction tracking both value and index
// Use cases: Global argmax computation
@group(0) @binding(0) var<storage, read> input: array<f64>;
@group(0) @binding(1) var<storage, read_write> output: array<u32>; // Partial results (indices)
struct Params {
size: u32,
}
@group(0) @binding(2) var<uniform> params: Params;
// Shared memory for values and indices
var<workgroup> shared_values: array<f64, 256>;
var<workgroup> shared_indices: array<u32, 256>;
@compute @workgroup_size(256)
fn main(
@builtin(global_invocation_id) global_id: vec3<u32>,
@builtin(local_invocation_id) local_id: vec3<u32>,
@builtin(workgroup_id) workgroup_id: vec3<u32>,
) {
let tid = local_id.x;
let gid = global_id.x;
// Load data into shared memory
var value: f64;
var index: u32;
if (gid < params.size) {
value = input[gid];
index = gid;
} else {
// Initialize with -FLT_MAX for max reduction
value = -3.402823e+38;
index = 0xFFFFFFFFu; // Invalid index
}
shared_values[tid] = value;
shared_indices[tid] = index;
workgroupBarrier();
// Tree reduction in shared memory, tracking indices
for (var stride = 128u; stride > 0u; stride = stride / 2u) {
if (tid < stride && (gid + stride) < params.size) {
let a_val = shared_values[tid];
let b_val = shared_values[tid + stride];
let a_idx = shared_indices[tid];
let b_idx = shared_indices[tid + stride];
if (b_val > a_val) {
shared_values[tid] = b_val;
shared_indices[tid] = b_idx;
} else {
// Keep a_val and a_idx (already in place)
}
}
workgroupBarrier();
}
// Write partial result (index of max value in this workgroup)
if (tid == 0u) {
output[workgroup_id.x] = shared_indices[0];
}
}