/**
 * GPU Compute - Compute shader infrastructure for WebGPU
 *
 * Provides GPU-accelerated data analysis capabilities.
 */
export declare const STATS_COMPUTE_WGSL = "\nstruct StatsResult {\n  min_val: f32,\n  max_val: f32,\n  sum: f32,\n  sum_sq: f32,\n  count: u32,\n  padding: vec3<u32>,\n}\n\n@group(0) @binding(0) var<storage, read> input_data: array<f32>;\n@group(0) @binding(1) var<storage, read_write> result: StatsResult;\n\nvar<workgroup> local_min: array<f32, 256>;\nvar<workgroup> local_max: array<f32, 256>;\nvar<workgroup> local_sum: array<f32, 256>;\nvar<workgroup> local_sum_sq: array<f32, 256>;\nvar<workgroup> local_count: array<u32, 256>;\n\n@compute @workgroup_size(256)\nfn main(\n  @builtin(global_invocation_id) global_id: vec3<u32>,\n  @builtin(local_invocation_id) local_id: vec3<u32>,\n  @builtin(workgroup_id) workgroup_id: vec3<u32>\n) {\n  let idx = global_id.x;\n  let lid = local_id.x;\n  let data_len = arrayLength(&input_data);\n  \n  // Initialize local values\n  if (idx < data_len) {\n    let val = input_data[idx];\n    local_min[lid] = val;\n    local_max[lid] = val;\n    local_sum[lid] = val;\n    local_sum_sq[lid] = val * val;\n    local_count[lid] = 1u;\n  } else {\n    local_min[lid] = 3.402823e+38; // f32 max\n    local_max[lid] = -3.402823e+38; // f32 min\n    local_sum[lid] = 0.0;\n    local_sum_sq[lid] = 0.0;\n    local_count[lid] = 0u;\n  }\n  \n  workgroupBarrier();\n  \n  // Parallel reduction\n  for (var stride: u32 = 128u; stride > 0u; stride = stride >> 1u) {\n    if (lid < stride) {\n      local_min[lid] = min(local_min[lid], local_min[lid + stride]);\n      local_max[lid] = max(local_max[lid], local_max[lid + stride]);\n      local_sum[lid] = local_sum[lid] + local_sum[lid + stride];\n      local_sum_sq[lid] = local_sum_sq[lid] + local_sum_sq[lid + stride];\n      local_count[lid] = local_count[lid] + local_count[lid + stride];\n    }\n    workgroupBarrier();\n  }\n  \n  // First thread writes result (atomic for multi-workgroup)\n  if (lid == 0u) {\n    // Use atomics for multi-workgroup reduction\n    // For now, we assume single workgroup or post-process on CPU\n    result.min_val = local_min[0];\n    result.max_val = local_max[0];\n    result.sum = local_sum[0];\n    result.sum_sq = local_sum_sq[0];\n    result.count = local_count[0];\n  }\n}\n";
export declare const MINMAX_COMPUTE_WGSL = "\nstruct MinMax {\n  min_x: f32,\n  max_x: f32,\n  min_y: f32,\n  max_y: f32,\n}\n\n@group(0) @binding(0) var<storage, read> points: array<vec2<f32>>;\n@group(0) @binding(1) var<storage, read_write> result: MinMax;\n\nvar<workgroup> local_min_x: array<f32, 256>;\nvar<workgroup> local_max_x: array<f32, 256>;\nvar<workgroup> local_min_y: array<f32, 256>;\nvar<workgroup> local_max_y: array<f32, 256>;\n\n@compute @workgroup_size(256)\nfn main(\n  @builtin(global_invocation_id) global_id: vec3<u32>,\n  @builtin(local_invocation_id) local_id: vec3<u32>\n) {\n  let idx = global_id.x;\n  let lid = local_id.x;\n  let data_len = arrayLength(&points);\n  \n  if (idx < data_len) {\n    let pt = points[idx];\n    local_min_x[lid] = pt.x;\n    local_max_x[lid] = pt.x;\n    local_min_y[lid] = pt.y;\n    local_max_y[lid] = pt.y;\n  } else {\n    local_min_x[lid] = 3.402823e+38;\n    local_max_x[lid] = -3.402823e+38;\n    local_min_y[lid] = 3.402823e+38;\n    local_max_y[lid] = -3.402823e+38;\n  }\n  \n  workgroupBarrier();\n  \n  for (var stride: u32 = 128u; stride > 0u; stride = stride >> 1u) {\n    if (lid < stride) {\n      local_min_x[lid] = min(local_min_x[lid], local_min_x[lid + stride]);\n      local_max_x[lid] = max(local_max_x[lid], local_max_x[lid + stride]);\n      local_min_y[lid] = min(local_min_y[lid], local_min_y[lid + stride]);\n      local_max_y[lid] = max(local_max_y[lid], local_max_y[lid + stride]);\n    }\n    workgroupBarrier();\n  }\n  \n  if (lid == 0u) {\n    result.min_x = local_min_x[0];\n    result.max_x = local_max_x[0];\n    result.min_y = local_min_y[0];\n    result.max_y = local_max_y[0];\n  }\n}\n";
export declare const DOWNSAMPLE_COMPUTE_WGSL = "\nstruct Params {\n  input_count: u32,\n  output_count: u32,\n  bucket_size: u32,\n  padding: u32,\n}\n\n@group(0) @binding(0) var<uniform> params: Params;\n@group(0) @binding(1) var<storage, read> input_points: array<vec2<f32>>;\n@group(0) @binding(2) var<storage, read_write> output_points: array<vec2<f32>>;\n\n@compute @workgroup_size(64)\nfn main(@builtin(global_invocation_id) global_id: vec3<u32>) {\n  let bucket_idx = global_id.x;\n  \n  if (bucket_idx >= params.output_count / 2u) {\n    return;\n  }\n  \n  let start = bucket_idx * params.bucket_size;\n  let end = min(start + params.bucket_size, params.input_count);\n  \n  if (start >= params.input_count) {\n    return;\n  }\n  \n  var min_y = input_points[start].y;\n  var max_y = input_points[start].y;\n  var min_idx = start;\n  var max_idx = start;\n  \n  for (var i = start; i < end; i = i + 1u) {\n    let y = input_points[i].y;\n    if (y < min_y) {\n      min_y = y;\n      min_idx = i;\n    }\n    if (y > max_y) {\n      max_y = y;\n      max_idx = i;\n    }\n  }\n  \n  // Output min and max points (preserving x,y pairs)\n  let out_base = bucket_idx * 2u;\n  \n  // Ensure min comes before max in x order\n  if (min_idx <= max_idx) {\n    output_points[out_base] = input_points[min_idx];\n    output_points[out_base + 1u] = input_points[max_idx];\n  } else {\n    output_points[out_base] = input_points[max_idx];\n    output_points[out_base + 1u] = input_points[min_idx];\n  }\n}\n";
export declare const PEAK_DETECT_COMPUTE_WGSL = "\nstruct Params {\n  data_count: u32,\n  threshold: f32,\n  min_distance: u32,\n  padding: u32,\n}\n\nstruct Peak {\n  index: u32,\n  value: f32,\n  is_peak: u32,\n  padding: u32,\n}\n\n@group(0) @binding(0) var<uniform> params: Params;\n@group(0) @binding(1) var<storage, read> data: array<f32>;\n@group(0) @binding(2) var<storage, read_write> peaks: array<Peak>;\n\n@compute @workgroup_size(64)\nfn main(@builtin(global_invocation_id) global_id: vec3<u32>) {\n  let idx = global_id.x;\n  \n  if (idx >= params.data_count) {\n    return;\n  }\n  \n  peaks[idx].index = idx;\n  peaks[idx].value = data[idx];\n  peaks[idx].is_peak = 0u;\n  \n  // Skip edges\n  if (idx < params.min_distance || idx >= params.data_count - params.min_distance) {\n    return;\n  }\n  \n  let val = data[idx];\n  \n  // Check if above threshold\n  if (val < params.threshold) {\n    return;\n  }\n  \n  // Check if local maximum\n  var is_max = true;\n  for (var i = 1u; i <= params.min_distance; i = i + 1u) {\n    if (data[idx - i] >= val || data[idx + i] >= val) {\n      is_max = false;\n      break;\n    }\n  }\n  \n  if (is_max) {\n    peaks[idx].is_peak = 1u;\n  }\n}\n";