| 1 | use crate::report::{make_filename_safe, BenchmarkId, MeasurementData, Report, ReportContext}; |
| 2 | use crate::stats::bivariate::regression::Slope; |
| 3 | |
| 4 | use crate::estimate::Estimate; |
| 5 | use crate::format; |
| 6 | use crate::fs; |
| 7 | use crate::measurement::ValueFormatter; |
| 8 | use crate::plot::{PlotContext, PlotData, Plotter}; |
| 9 | use crate::SavedSample; |
| 10 | use criterion_plot::Size; |
| 11 | use serde::Serialize; |
| 12 | use std::cell::RefCell; |
| 13 | use std::cmp::Ordering; |
| 14 | use std::collections::{BTreeSet, HashMap}; |
| 15 | use std::path::{Path, PathBuf}; |
| 16 | use tinytemplate::TinyTemplate; |
| 17 | |
| 18 | const THUMBNAIL_SIZE: Option<Size> = Some(Size(450, 300)); |
| 19 | |
| 20 | fn debug_context<S: Serialize>(path: &Path, context: &S) { |
| 21 | if crate::debug_enabled() { |
| 22 | let mut context_path = PathBuf::from(path); |
| 23 | context_path.set_extension("json" ); |
| 24 | println!("Writing report context to {:?}" , context_path); |
| 25 | let result = fs::save(context, &context_path); |
| 26 | if let Err(e) = result { |
| 27 | error!("Failed to write report context debug output: {}" , e); |
| 28 | } |
| 29 | } |
| 30 | } |
| 31 | |
| 32 | #[derive(Serialize)] |
| 33 | struct Context { |
| 34 | title: String, |
| 35 | confidence: String, |
| 36 | |
| 37 | thumbnail_width: usize, |
| 38 | thumbnail_height: usize, |
| 39 | |
| 40 | slope: Option<ConfidenceInterval>, |
| 41 | r2: ConfidenceInterval, |
| 42 | mean: ConfidenceInterval, |
| 43 | std_dev: ConfidenceInterval, |
| 44 | median: ConfidenceInterval, |
| 45 | mad: ConfidenceInterval, |
| 46 | throughput: Option<ConfidenceInterval>, |
| 47 | |
| 48 | additional_plots: Vec<Plot>, |
| 49 | |
| 50 | comparison: Option<Comparison>, |
| 51 | } |
| 52 | |
| 53 | #[derive(Serialize)] |
| 54 | struct IndividualBenchmark { |
| 55 | name: String, |
| 56 | path: String, |
| 57 | regression_exists: bool, |
| 58 | } |
| 59 | impl IndividualBenchmark { |
| 60 | fn from_id( |
| 61 | output_directory: &Path, |
| 62 | path_prefix: &str, |
| 63 | id: &BenchmarkId, |
| 64 | ) -> IndividualBenchmark { |
| 65 | let mut regression_path = PathBuf::from(output_directory); |
| 66 | regression_path.push(id.as_directory_name()); |
| 67 | regression_path.push("report" ); |
| 68 | regression_path.push("regression.svg" ); |
| 69 | |
| 70 | IndividualBenchmark { |
| 71 | name: id.as_title().to_owned(), |
| 72 | path: format!("{}/{}" , path_prefix, id.as_directory_name()), |
| 73 | regression_exists: regression_path.is_file(), |
| 74 | } |
| 75 | } |
| 76 | } |
| 77 | |
| 78 | #[derive(Serialize)] |
| 79 | struct SummaryContext { |
| 80 | group_id: String, |
| 81 | |
| 82 | thumbnail_width: usize, |
| 83 | thumbnail_height: usize, |
| 84 | |
| 85 | violin_plot: Option<String>, |
| 86 | line_chart: Option<String>, |
| 87 | |
| 88 | benchmarks: Vec<IndividualBenchmark>, |
| 89 | } |
| 90 | |
| 91 | #[derive(Serialize)] |
| 92 | struct ConfidenceInterval { |
| 93 | lower: String, |
| 94 | upper: String, |
| 95 | point: String, |
| 96 | } |
| 97 | |
| 98 | #[derive(Serialize)] |
| 99 | struct Plot { |
| 100 | name: String, |
| 101 | url: String, |
| 102 | } |
| 103 | impl Plot { |
| 104 | fn new(name: &str, url: &str) -> Plot { |
| 105 | Plot { |
| 106 | name: name.to_owned(), |
| 107 | url: url.to_owned(), |
| 108 | } |
| 109 | } |
| 110 | } |
| 111 | |
| 112 | #[derive(Serialize)] |
| 113 | struct Comparison { |
| 114 | p_value: String, |
| 115 | inequality: String, |
| 116 | significance_level: String, |
| 117 | explanation: String, |
| 118 | |
| 119 | change: ConfidenceInterval, |
| 120 | thrpt_change: Option<ConfidenceInterval>, |
| 121 | additional_plots: Vec<Plot>, |
| 122 | } |
| 123 | |
| 124 | fn if_exists(output_directory: &Path, path: &Path) -> Option<String> { |
| 125 | let report_path = path.join("report/index.html" ); |
| 126 | if PathBuf::from(output_directory).join(&report_path).is_file() { |
| 127 | Some(report_path.to_string_lossy().to_string()) |
| 128 | } else { |
| 129 | None |
| 130 | } |
| 131 | } |
| 132 | #[derive(Serialize, Debug)] |
| 133 | struct ReportLink<'a> { |
| 134 | name: &'a str, |
| 135 | path: Option<String>, |
| 136 | } |
| 137 | impl<'a> ReportLink<'a> { |
| 138 | // TODO: Would be nice if I didn't have to keep making these components filename-safe. |
| 139 | fn group(output_directory: &Path, group_id: &'a str) -> ReportLink<'a> { |
| 140 | let path = PathBuf::from(make_filename_safe(group_id)); |
| 141 | |
| 142 | ReportLink { |
| 143 | name: group_id, |
| 144 | path: if_exists(output_directory, &path), |
| 145 | } |
| 146 | } |
| 147 | |
| 148 | fn function(output_directory: &Path, group_id: &str, function_id: &'a str) -> ReportLink<'a> { |
| 149 | let mut path = PathBuf::from(make_filename_safe(group_id)); |
| 150 | path.push(make_filename_safe(function_id)); |
| 151 | |
| 152 | ReportLink { |
| 153 | name: function_id, |
| 154 | path: if_exists(output_directory, &path), |
| 155 | } |
| 156 | } |
| 157 | |
| 158 | fn value(output_directory: &Path, group_id: &str, value_str: &'a str) -> ReportLink<'a> { |
| 159 | let mut path = PathBuf::from(make_filename_safe(group_id)); |
| 160 | path.push(make_filename_safe(value_str)); |
| 161 | |
| 162 | ReportLink { |
| 163 | name: value_str, |
| 164 | path: if_exists(output_directory, &path), |
| 165 | } |
| 166 | } |
| 167 | |
| 168 | fn individual(output_directory: &Path, id: &'a BenchmarkId) -> ReportLink<'a> { |
| 169 | let path = PathBuf::from(id.as_directory_name()); |
| 170 | ReportLink { |
| 171 | name: id.as_title(), |
| 172 | path: if_exists(output_directory, &path), |
| 173 | } |
| 174 | } |
| 175 | } |
| 176 | |
| 177 | #[derive(Serialize)] |
| 178 | struct BenchmarkValueGroup<'a> { |
| 179 | value: Option<ReportLink<'a>>, |
| 180 | benchmarks: Vec<ReportLink<'a>>, |
| 181 | } |
| 182 | |
| 183 | #[derive(Serialize)] |
| 184 | struct BenchmarkGroup<'a> { |
| 185 | group_report: ReportLink<'a>, |
| 186 | |
| 187 | function_ids: Option<Vec<ReportLink<'a>>>, |
| 188 | values: Option<Vec<ReportLink<'a>>>, |
| 189 | |
| 190 | individual_links: Vec<BenchmarkValueGroup<'a>>, |
| 191 | } |
| 192 | impl<'a> BenchmarkGroup<'a> { |
| 193 | fn new(output_directory: &Path, ids: &[&'a BenchmarkId]) -> BenchmarkGroup<'a> { |
| 194 | let group_id = &ids[0].group_id; |
| 195 | let group_report = ReportLink::group(output_directory, group_id); |
| 196 | |
| 197 | let mut function_ids = Vec::with_capacity(ids.len()); |
| 198 | let mut values = Vec::with_capacity(ids.len()); |
| 199 | let mut individual_links = HashMap::with_capacity(ids.len()); |
| 200 | |
| 201 | for id in ids.iter() { |
| 202 | let function_id = id.function_id.as_deref(); |
| 203 | let value = id.value_str.as_deref(); |
| 204 | |
| 205 | let individual_link = ReportLink::individual(output_directory, id); |
| 206 | |
| 207 | function_ids.push(function_id); |
| 208 | values.push(value); |
| 209 | |
| 210 | individual_links.insert((function_id, value), individual_link); |
| 211 | } |
| 212 | |
| 213 | fn parse_opt(os: &Option<&str>) -> Option<f64> { |
| 214 | os.and_then(|s| s.parse::<f64>().ok()) |
| 215 | } |
| 216 | |
| 217 | // If all of the value strings can be parsed into a number, sort/dedupe |
| 218 | // numerically. Otherwise sort lexicographically. |
| 219 | if values.iter().all(|os| parse_opt(os).is_some()) { |
| 220 | values.sort_unstable_by(|v1, v2| { |
| 221 | let num1 = parse_opt(v1); |
| 222 | let num2 = parse_opt(v2); |
| 223 | |
| 224 | num1.partial_cmp(&num2).unwrap_or(Ordering::Less) |
| 225 | }); |
| 226 | values.dedup_by_key(|os| parse_opt(os).unwrap()); |
| 227 | } else { |
| 228 | values.sort_unstable(); |
| 229 | values.dedup(); |
| 230 | } |
| 231 | |
| 232 | // Sort and dedupe functions by name. |
| 233 | function_ids.sort_unstable(); |
| 234 | function_ids.dedup(); |
| 235 | |
| 236 | let mut value_groups = Vec::with_capacity(values.len()); |
| 237 | for value in values.iter() { |
| 238 | let row = function_ids |
| 239 | .iter() |
| 240 | .filter_map(|f| individual_links.remove(&(*f, *value))) |
| 241 | .collect::<Vec<_>>(); |
| 242 | value_groups.push(BenchmarkValueGroup { |
| 243 | value: value.map(|s| ReportLink::value(output_directory, group_id, s)), |
| 244 | benchmarks: row, |
| 245 | }); |
| 246 | } |
| 247 | |
| 248 | let function_ids = function_ids |
| 249 | .into_iter() |
| 250 | .map(|os| os.map(|s| ReportLink::function(output_directory, group_id, s))) |
| 251 | .collect::<Option<Vec<_>>>(); |
| 252 | let values = values |
| 253 | .into_iter() |
| 254 | .map(|os| os.map(|s| ReportLink::value(output_directory, group_id, s))) |
| 255 | .collect::<Option<Vec<_>>>(); |
| 256 | |
| 257 | BenchmarkGroup { |
| 258 | group_report, |
| 259 | function_ids, |
| 260 | values, |
| 261 | individual_links: value_groups, |
| 262 | } |
| 263 | } |
| 264 | } |
| 265 | |
| 266 | #[derive(Serialize)] |
| 267 | struct IndexContext<'a> { |
| 268 | groups: Vec<BenchmarkGroup<'a>>, |
| 269 | } |
| 270 | |
| 271 | pub struct Html { |
| 272 | templates: TinyTemplate<'static>, |
| 273 | plotter: RefCell<Box<dyn Plotter>>, |
| 274 | } |
| 275 | impl Html { |
| 276 | pub(crate) fn new(plotter: Box<dyn Plotter>) -> Html { |
| 277 | let mut templates = TinyTemplate::new(); |
| 278 | templates |
| 279 | .add_template("report_link" , include_str!("report_link.html.tt" )) |
| 280 | .expect("Unable to parse report_link template." ); |
| 281 | templates |
| 282 | .add_template("index" , include_str!("index.html.tt" )) |
| 283 | .expect("Unable to parse index template." ); |
| 284 | templates |
| 285 | .add_template("benchmark_report" , include_str!("benchmark_report.html.tt" )) |
| 286 | .expect("Unable to parse benchmark_report template" ); |
| 287 | templates |
| 288 | .add_template("summary_report" , include_str!("summary_report.html.tt" )) |
| 289 | .expect("Unable to parse summary_report template" ); |
| 290 | |
| 291 | let plotter = RefCell::new(plotter); |
| 292 | Html { templates, plotter } |
| 293 | } |
| 294 | } |
| 295 | impl Report for Html { |
| 296 | fn measurement_complete( |
| 297 | &self, |
| 298 | id: &BenchmarkId, |
| 299 | report_context: &ReportContext, |
| 300 | measurements: &MeasurementData<'_>, |
| 301 | formatter: &dyn ValueFormatter, |
| 302 | ) { |
| 303 | try_else_return!({ |
| 304 | let mut report_dir = report_context.output_directory.clone(); |
| 305 | report_dir.push(id.as_directory_name()); |
| 306 | report_dir.push("report" ); |
| 307 | fs::mkdirp(&report_dir) |
| 308 | }); |
| 309 | |
| 310 | let typical_estimate = &measurements.absolute_estimates.typical(); |
| 311 | |
| 312 | let time_interval = |est: &Estimate| -> ConfidenceInterval { |
| 313 | ConfidenceInterval { |
| 314 | lower: formatter.format_value(est.confidence_interval.lower_bound), |
| 315 | point: formatter.format_value(est.point_estimate), |
| 316 | upper: formatter.format_value(est.confidence_interval.upper_bound), |
| 317 | } |
| 318 | }; |
| 319 | |
| 320 | let data = measurements.data; |
| 321 | |
| 322 | elapsed! { |
| 323 | "Generating plots" , |
| 324 | self.generate_plots(id, report_context, formatter, measurements) |
| 325 | } |
| 326 | |
| 327 | let mut additional_plots = vec![ |
| 328 | Plot::new("Typical" , "typical.svg" ), |
| 329 | Plot::new("Mean" , "mean.svg" ), |
| 330 | Plot::new("Std. Dev." , "SD.svg" ), |
| 331 | Plot::new("Median" , "median.svg" ), |
| 332 | Plot::new("MAD" , "MAD.svg" ), |
| 333 | ]; |
| 334 | if measurements.absolute_estimates.slope.is_some() { |
| 335 | additional_plots.push(Plot::new("Slope" , "slope.svg" )); |
| 336 | } |
| 337 | |
| 338 | let throughput = measurements |
| 339 | .throughput |
| 340 | .as_ref() |
| 341 | .map(|thr| ConfidenceInterval { |
| 342 | lower: formatter |
| 343 | .format_throughput(thr, typical_estimate.confidence_interval.upper_bound), |
| 344 | upper: formatter |
| 345 | .format_throughput(thr, typical_estimate.confidence_interval.lower_bound), |
| 346 | point: formatter.format_throughput(thr, typical_estimate.point_estimate), |
| 347 | }); |
| 348 | |
| 349 | let context = Context { |
| 350 | title: id.as_title().to_owned(), |
| 351 | confidence: format!( |
| 352 | "{:.2}" , |
| 353 | typical_estimate.confidence_interval.confidence_level |
| 354 | ), |
| 355 | |
| 356 | thumbnail_width: THUMBNAIL_SIZE.unwrap().0, |
| 357 | thumbnail_height: THUMBNAIL_SIZE.unwrap().1, |
| 358 | |
| 359 | slope: measurements |
| 360 | .absolute_estimates |
| 361 | .slope |
| 362 | .as_ref() |
| 363 | .map(time_interval), |
| 364 | mean: time_interval(&measurements.absolute_estimates.mean), |
| 365 | median: time_interval(&measurements.absolute_estimates.median), |
| 366 | mad: time_interval(&measurements.absolute_estimates.median_abs_dev), |
| 367 | std_dev: time_interval(&measurements.absolute_estimates.std_dev), |
| 368 | throughput, |
| 369 | |
| 370 | r2: ConfidenceInterval { |
| 371 | lower: format!( |
| 372 | "{:0.7}" , |
| 373 | Slope(typical_estimate.confidence_interval.lower_bound).r_squared(&data) |
| 374 | ), |
| 375 | upper: format!( |
| 376 | "{:0.7}" , |
| 377 | Slope(typical_estimate.confidence_interval.upper_bound).r_squared(&data) |
| 378 | ), |
| 379 | point: format!( |
| 380 | "{:0.7}" , |
| 381 | Slope(typical_estimate.point_estimate).r_squared(&data) |
| 382 | ), |
| 383 | }, |
| 384 | |
| 385 | additional_plots, |
| 386 | |
| 387 | comparison: self.comparison(measurements), |
| 388 | }; |
| 389 | |
| 390 | let mut report_path = report_context.output_directory.clone(); |
| 391 | report_path.push(id.as_directory_name()); |
| 392 | report_path.push("report" ); |
| 393 | report_path.push("index.html" ); |
| 394 | debug_context(&report_path, &context); |
| 395 | |
| 396 | let text = self |
| 397 | .templates |
| 398 | .render("benchmark_report" , &context) |
| 399 | .expect("Failed to render benchmark report template" ); |
| 400 | try_else_return!(fs::save_string(&text, &report_path)); |
| 401 | } |
| 402 | |
| 403 | fn summarize( |
| 404 | &self, |
| 405 | context: &ReportContext, |
| 406 | all_ids: &[BenchmarkId], |
| 407 | formatter: &dyn ValueFormatter, |
| 408 | ) { |
| 409 | let all_ids = all_ids |
| 410 | .iter() |
| 411 | .filter(|id| { |
| 412 | let id_dir = context.output_directory.join(id.as_directory_name()); |
| 413 | fs::is_dir(&id_dir) |
| 414 | }) |
| 415 | .collect::<Vec<_>>(); |
| 416 | if all_ids.is_empty() { |
| 417 | return; |
| 418 | } |
| 419 | |
| 420 | let group_id = all_ids[0].group_id.clone(); |
| 421 | |
| 422 | let data = self.load_summary_data(&context.output_directory, &all_ids); |
| 423 | |
| 424 | let mut function_ids = BTreeSet::new(); |
| 425 | let mut value_strs = Vec::with_capacity(all_ids.len()); |
| 426 | for id in all_ids { |
| 427 | if let Some(ref function_id) = id.function_id { |
| 428 | function_ids.insert(function_id); |
| 429 | } |
| 430 | if let Some(ref value_str) = id.value_str { |
| 431 | value_strs.push(value_str); |
| 432 | } |
| 433 | } |
| 434 | |
| 435 | fn try_parse(s: &str) -> Option<f64> { |
| 436 | s.parse::<f64>().ok() |
| 437 | } |
| 438 | |
| 439 | // If all of the value strings can be parsed into a number, sort/dedupe |
| 440 | // numerically. Otherwise sort lexicographically. |
| 441 | if value_strs.iter().all(|os| try_parse(os).is_some()) { |
| 442 | value_strs.sort_unstable_by(|v1, v2| { |
| 443 | let num1 = try_parse(v1); |
| 444 | let num2 = try_parse(v2); |
| 445 | |
| 446 | num1.partial_cmp(&num2).unwrap_or(Ordering::Less) |
| 447 | }); |
| 448 | value_strs.dedup_by_key(|os| try_parse(os).unwrap()); |
| 449 | } else { |
| 450 | value_strs.sort_unstable(); |
| 451 | value_strs.dedup(); |
| 452 | } |
| 453 | |
| 454 | for function_id in function_ids { |
| 455 | let samples_with_function: Vec<_> = data |
| 456 | .iter() |
| 457 | .by_ref() |
| 458 | .filter(|&&(id, _)| id.function_id.as_ref() == Some(function_id)) |
| 459 | .collect(); |
| 460 | |
| 461 | if samples_with_function.len() > 1 { |
| 462 | let subgroup_id = |
| 463 | BenchmarkId::new(group_id.clone(), Some(function_id.clone()), None, None); |
| 464 | |
| 465 | self.generate_summary( |
| 466 | &subgroup_id, |
| 467 | &samples_with_function, |
| 468 | context, |
| 469 | formatter, |
| 470 | false, |
| 471 | ); |
| 472 | } |
| 473 | } |
| 474 | |
| 475 | for value_str in value_strs { |
| 476 | let samples_with_value: Vec<_> = data |
| 477 | .iter() |
| 478 | .by_ref() |
| 479 | .filter(|&&(id, _)| id.value_str.as_ref() == Some(value_str)) |
| 480 | .collect(); |
| 481 | |
| 482 | if samples_with_value.len() > 1 { |
| 483 | let subgroup_id = |
| 484 | BenchmarkId::new(group_id.clone(), None, Some(value_str.clone()), None); |
| 485 | |
| 486 | self.generate_summary(&subgroup_id, &samples_with_value, context, formatter, false); |
| 487 | } |
| 488 | } |
| 489 | |
| 490 | let mut all_data = data.iter().by_ref().collect::<Vec<_>>(); |
| 491 | // First sort the ids/data by value. |
| 492 | // If all of the value strings can be parsed into a number, sort/dedupe |
| 493 | // numerically. Otherwise sort lexicographically. |
| 494 | let all_values_numeric = all_data |
| 495 | .iter() |
| 496 | .all(|(id, _)| id.value_str.as_deref().and_then(try_parse).is_some()); |
| 497 | if all_values_numeric { |
| 498 | all_data.sort_unstable_by(|(a, _), (b, _)| { |
| 499 | let num1 = a.value_str.as_deref().and_then(try_parse); |
| 500 | let num2 = b.value_str.as_deref().and_then(try_parse); |
| 501 | |
| 502 | num1.partial_cmp(&num2).unwrap_or(Ordering::Less) |
| 503 | }); |
| 504 | } else { |
| 505 | all_data.sort_unstable_by_key(|(id, _)| id.value_str.as_ref()); |
| 506 | } |
| 507 | // Next, sort the ids/data by function name. This results in a sorting priority of |
| 508 | // function name, then value. This one has to be a stable sort. |
| 509 | all_data.sort_by_key(|(id, _)| id.function_id.as_ref()); |
| 510 | |
| 511 | self.generate_summary( |
| 512 | &BenchmarkId::new(group_id, None, None, None), |
| 513 | &all_data, |
| 514 | context, |
| 515 | formatter, |
| 516 | true, |
| 517 | ); |
| 518 | self.plotter.borrow_mut().wait(); |
| 519 | } |
| 520 | |
| 521 | fn final_summary(&self, report_context: &ReportContext) { |
| 522 | let output_directory = &report_context.output_directory; |
| 523 | if !fs::is_dir(&output_directory) { |
| 524 | return; |
| 525 | } |
| 526 | |
| 527 | let mut found_ids = try_else_return!(fs::list_existing_benchmarks(&output_directory)); |
| 528 | found_ids.sort_unstable_by_key(|id| id.id().to_owned()); |
| 529 | |
| 530 | // Group IDs by group id |
| 531 | let mut id_groups: HashMap<&str, Vec<&BenchmarkId>> = HashMap::new(); |
| 532 | for id in found_ids.iter() { |
| 533 | id_groups |
| 534 | .entry(&id.group_id) |
| 535 | .or_insert_with(Vec::new) |
| 536 | .push(id); |
| 537 | } |
| 538 | |
| 539 | let mut groups = id_groups |
| 540 | .into_values() |
| 541 | .map(|group| BenchmarkGroup::new(output_directory, &group)) |
| 542 | .collect::<Vec<BenchmarkGroup<'_>>>(); |
| 543 | groups.sort_unstable_by_key(|g| g.group_report.name); |
| 544 | |
| 545 | try_else_return!(fs::mkdirp(&output_directory.join("report" ))); |
| 546 | |
| 547 | let report_path = output_directory.join("report" ).join("index.html" ); |
| 548 | |
| 549 | let context = IndexContext { groups }; |
| 550 | |
| 551 | debug_context(&report_path, &context); |
| 552 | |
| 553 | let text = self |
| 554 | .templates |
| 555 | .render("index" , &context) |
| 556 | .expect("Failed to render index template" ); |
| 557 | try_else_return!(fs::save_string(&text, &report_path,)); |
| 558 | } |
| 559 | } |
| 560 | impl Html { |
| 561 | fn comparison(&self, measurements: &MeasurementData<'_>) -> Option<Comparison> { |
| 562 | if let Some(ref comp) = measurements.comparison { |
| 563 | let different_mean = comp.p_value < comp.significance_threshold; |
| 564 | let mean_est = &comp.relative_estimates.mean; |
| 565 | let explanation_str: String; |
| 566 | |
| 567 | if !different_mean { |
| 568 | explanation_str = "No change in performance detected." .to_owned(); |
| 569 | } else { |
| 570 | let comparison = compare_to_threshold(mean_est, comp.noise_threshold); |
| 571 | match comparison { |
| 572 | ComparisonResult::Improved => { |
| 573 | explanation_str = "Performance has improved." .to_owned(); |
| 574 | } |
| 575 | ComparisonResult::Regressed => { |
| 576 | explanation_str = "Performance has regressed." .to_owned(); |
| 577 | } |
| 578 | ComparisonResult::NonSignificant => { |
| 579 | explanation_str = "Change within noise threshold." .to_owned(); |
| 580 | } |
| 581 | } |
| 582 | } |
| 583 | |
| 584 | let comp = Comparison { |
| 585 | p_value: format!("{:.2}" , comp.p_value), |
| 586 | inequality: (if different_mean { "<" } else { ">" }).to_owned(), |
| 587 | significance_level: format!("{:.2}" , comp.significance_threshold), |
| 588 | explanation: explanation_str, |
| 589 | |
| 590 | change: ConfidenceInterval { |
| 591 | point: format::change(mean_est.point_estimate, true), |
| 592 | lower: format::change(mean_est.confidence_interval.lower_bound, true), |
| 593 | upper: format::change(mean_est.confidence_interval.upper_bound, true), |
| 594 | }, |
| 595 | |
| 596 | thrpt_change: measurements.throughput.as_ref().map(|_| { |
| 597 | let to_thrpt_estimate = |ratio: f64| 1.0 / (1.0 + ratio) - 1.0; |
| 598 | ConfidenceInterval { |
| 599 | point: format::change(to_thrpt_estimate(mean_est.point_estimate), true), |
| 600 | lower: format::change( |
| 601 | to_thrpt_estimate(mean_est.confidence_interval.lower_bound), |
| 602 | true, |
| 603 | ), |
| 604 | upper: format::change( |
| 605 | to_thrpt_estimate(mean_est.confidence_interval.upper_bound), |
| 606 | true, |
| 607 | ), |
| 608 | } |
| 609 | }), |
| 610 | |
| 611 | additional_plots: vec![ |
| 612 | Plot::new("Change in mean" , "change/mean.svg" ), |
| 613 | Plot::new("Change in median" , "change/median.svg" ), |
| 614 | Plot::new("T-Test" , "change/t-test.svg" ), |
| 615 | ], |
| 616 | }; |
| 617 | Some(comp) |
| 618 | } else { |
| 619 | None |
| 620 | } |
| 621 | } |
| 622 | |
| 623 | fn generate_plots( |
| 624 | &self, |
| 625 | id: &BenchmarkId, |
| 626 | context: &ReportContext, |
| 627 | formatter: &dyn ValueFormatter, |
| 628 | measurements: &MeasurementData<'_>, |
| 629 | ) { |
| 630 | let plot_ctx = PlotContext { |
| 631 | id, |
| 632 | context, |
| 633 | size: None, |
| 634 | is_thumbnail: false, |
| 635 | }; |
| 636 | |
| 637 | let plot_data = PlotData { |
| 638 | measurements, |
| 639 | formatter, |
| 640 | comparison: None, |
| 641 | }; |
| 642 | |
| 643 | let plot_ctx_small = plot_ctx.thumbnail(true).size(THUMBNAIL_SIZE); |
| 644 | |
| 645 | self.plotter.borrow_mut().pdf(plot_ctx, plot_data); |
| 646 | self.plotter.borrow_mut().pdf(plot_ctx_small, plot_data); |
| 647 | if measurements.absolute_estimates.slope.is_some() { |
| 648 | self.plotter.borrow_mut().regression(plot_ctx, plot_data); |
| 649 | self.plotter |
| 650 | .borrow_mut() |
| 651 | .regression(plot_ctx_small, plot_data); |
| 652 | } else { |
| 653 | self.plotter |
| 654 | .borrow_mut() |
| 655 | .iteration_times(plot_ctx, plot_data); |
| 656 | self.plotter |
| 657 | .borrow_mut() |
| 658 | .iteration_times(plot_ctx_small, plot_data); |
| 659 | } |
| 660 | |
| 661 | self.plotter |
| 662 | .borrow_mut() |
| 663 | .abs_distributions(plot_ctx, plot_data); |
| 664 | |
| 665 | if let Some(ref comp) = measurements.comparison { |
| 666 | try_else_return!({ |
| 667 | let mut change_dir = context.output_directory.clone(); |
| 668 | change_dir.push(id.as_directory_name()); |
| 669 | change_dir.push("report" ); |
| 670 | change_dir.push("change" ); |
| 671 | fs::mkdirp(&change_dir) |
| 672 | }); |
| 673 | |
| 674 | try_else_return!({ |
| 675 | let mut both_dir = context.output_directory.clone(); |
| 676 | both_dir.push(id.as_directory_name()); |
| 677 | both_dir.push("report" ); |
| 678 | both_dir.push("both" ); |
| 679 | fs::mkdirp(&both_dir) |
| 680 | }); |
| 681 | |
| 682 | let comp_data = plot_data.comparison(comp); |
| 683 | |
| 684 | self.plotter.borrow_mut().pdf(plot_ctx, comp_data); |
| 685 | self.plotter.borrow_mut().pdf(plot_ctx_small, comp_data); |
| 686 | if measurements.absolute_estimates.slope.is_some() |
| 687 | && comp.base_estimates.slope.is_some() |
| 688 | { |
| 689 | self.plotter.borrow_mut().regression(plot_ctx, comp_data); |
| 690 | self.plotter |
| 691 | .borrow_mut() |
| 692 | .regression(plot_ctx_small, comp_data); |
| 693 | } else { |
| 694 | self.plotter |
| 695 | .borrow_mut() |
| 696 | .iteration_times(plot_ctx, comp_data); |
| 697 | self.plotter |
| 698 | .borrow_mut() |
| 699 | .iteration_times(plot_ctx_small, comp_data); |
| 700 | } |
| 701 | self.plotter.borrow_mut().t_test(plot_ctx, comp_data); |
| 702 | self.plotter |
| 703 | .borrow_mut() |
| 704 | .rel_distributions(plot_ctx, comp_data); |
| 705 | } |
| 706 | |
| 707 | self.plotter.borrow_mut().wait(); |
| 708 | } |
| 709 | |
| 710 | fn load_summary_data<'a>( |
| 711 | &self, |
| 712 | output_directory: &Path, |
| 713 | all_ids: &[&'a BenchmarkId], |
| 714 | ) -> Vec<(&'a BenchmarkId, Vec<f64>)> { |
| 715 | all_ids |
| 716 | .iter() |
| 717 | .filter_map(|id| { |
| 718 | let entry = output_directory.join(id.as_directory_name()).join("new" ); |
| 719 | |
| 720 | let SavedSample { iters, times, .. } = |
| 721 | try_else_return!(fs::load(&entry.join("sample.json" )), || None); |
| 722 | let avg_times = iters |
| 723 | .into_iter() |
| 724 | .zip(times.into_iter()) |
| 725 | .map(|(iters, time)| time / iters) |
| 726 | .collect::<Vec<_>>(); |
| 727 | |
| 728 | Some((*id, avg_times)) |
| 729 | }) |
| 730 | .collect::<Vec<_>>() |
| 731 | } |
| 732 | |
| 733 | fn generate_summary( |
| 734 | &self, |
| 735 | id: &BenchmarkId, |
| 736 | data: &[&(&BenchmarkId, Vec<f64>)], |
| 737 | report_context: &ReportContext, |
| 738 | formatter: &dyn ValueFormatter, |
| 739 | full_summary: bool, |
| 740 | ) { |
| 741 | let plot_ctx = PlotContext { |
| 742 | id, |
| 743 | context: report_context, |
| 744 | size: None, |
| 745 | is_thumbnail: false, |
| 746 | }; |
| 747 | |
| 748 | try_else_return!( |
| 749 | { |
| 750 | let mut report_dir = report_context.output_directory.clone(); |
| 751 | report_dir.push(id.as_directory_name()); |
| 752 | report_dir.push("report" ); |
| 753 | fs::mkdirp(&report_dir) |
| 754 | }, |
| 755 | || {} |
| 756 | ); |
| 757 | |
| 758 | self.plotter.borrow_mut().violin(plot_ctx, formatter, data); |
| 759 | |
| 760 | let value_types: Vec<_> = data.iter().map(|&&(id, _)| id.value_type()).collect(); |
| 761 | let mut line_path = None; |
| 762 | |
| 763 | if value_types.iter().all(|x| x == &value_types[0]) { |
| 764 | if let Some(value_type) = value_types[0] { |
| 765 | let values: Vec<_> = data.iter().map(|&&(id, _)| id.as_number()).collect(); |
| 766 | if values.iter().any(|x| x != &values[0]) { |
| 767 | self.plotter |
| 768 | .borrow_mut() |
| 769 | .line_comparison(plot_ctx, formatter, data, value_type); |
| 770 | line_path = Some(plot_ctx.line_comparison_path()); |
| 771 | } |
| 772 | } |
| 773 | } |
| 774 | |
| 775 | let path_prefix = if full_summary { "../.." } else { "../../.." }; |
| 776 | let benchmarks = data |
| 777 | .iter() |
| 778 | .map(|&&(id, _)| { |
| 779 | IndividualBenchmark::from_id(&report_context.output_directory, path_prefix, id) |
| 780 | }) |
| 781 | .collect(); |
| 782 | |
| 783 | let context = SummaryContext { |
| 784 | group_id: id.as_title().to_owned(), |
| 785 | |
| 786 | thumbnail_width: THUMBNAIL_SIZE.unwrap().0, |
| 787 | thumbnail_height: THUMBNAIL_SIZE.unwrap().1, |
| 788 | |
| 789 | violin_plot: Some(plot_ctx.violin_path().to_string_lossy().into_owned()), |
| 790 | line_chart: line_path.map(|p| p.to_string_lossy().into_owned()), |
| 791 | |
| 792 | benchmarks, |
| 793 | }; |
| 794 | |
| 795 | let mut report_path = report_context.output_directory.clone(); |
| 796 | report_path.push(id.as_directory_name()); |
| 797 | report_path.push("report" ); |
| 798 | report_path.push("index.html" ); |
| 799 | debug_context(&report_path, &context); |
| 800 | |
| 801 | let text = self |
| 802 | .templates |
| 803 | .render("summary_report" , &context) |
| 804 | .expect("Failed to render summary report template" ); |
| 805 | try_else_return!(fs::save_string(&text, &report_path,), || {}); |
| 806 | } |
| 807 | } |
| 808 | |
| 809 | enum ComparisonResult { |
| 810 | Improved, |
| 811 | Regressed, |
| 812 | NonSignificant, |
| 813 | } |
| 814 | |
| 815 | fn compare_to_threshold(estimate: &Estimate, noise: f64) -> ComparisonResult { |
| 816 | let ci = &estimate.confidence_interval; |
| 817 | let lb = ci.lower_bound; |
| 818 | let ub = ci.upper_bound; |
| 819 | |
| 820 | if lb < -noise && ub < -noise { |
| 821 | ComparisonResult::Improved |
| 822 | } else if lb > noise && ub > noise { |
| 823 | ComparisonResult::Regressed |
| 824 | } else { |
| 825 | ComparisonResult::NonSignificant |
| 826 | } |
| 827 | } |
| 828 | |