Files
wasmtime/lib/cretonne/src/timing.rs
Jakob Stoklund Olesen 91343f208d Fix quadratic behavior in sequence numbering.
The ir::layout module is assigning sequence numbers to all EBBs and
instructions so relative positions can be computed in constant time.
This works a lot like BASIC line numbers where we initially use numbers
10, 20, 30, ... so we can insert new instructions in the middle of the
sequence without renumbering everything.

In some cases where the coalescer is misbehaving and inserting a lot of
copy instructions, we end up having to renumber a larger and larger
number of instructions to make space in the sequence. This causes the
following reload pass to be very slow, spending most of its time
renumbering instructions.

Fix this by putting an upper limit on the number of instructions we're
willing to renumber locally. When the limit is reached, switch to a full
function renumbering with the major stride of 10. This gives us new
elasticity in the sequence numbers.

- Time to compile the Python interpreter in #229 drops from 4826 s -> 15.8 s.
- The godot benchmark in #226 drops from 1257 s -> 75 s.
- The AngryBots1 benchmark does not have the coalescer misbehavior.
  Its compilation time changes 22.9 s -> 23.1 s.

It's worth noting that the sequence numbering is still technically
quadratic with this fix. The system is not designed to handle a large
number of instructions inserted in a single location. It expects a more
even distribution of new instructions.

We still need to fix the coalescer. It should not insert so many copies
in degenerate cases.
2018-01-22 09:42:26 -08:00

227 lines
7.3 KiB
Rust

//! Pass timing.
//!
//! This modules provides facilities for timing the execution of individual compilation passes.
use std::fmt;
pub use self::details::{TimingToken, PassTimes, take_current, add_to_current};
// Each pass that can be timed is predefined with the `define_passes!` macro. Each pass has a
// snake_case name and a plain text description used when printing out the timing report.
//
// This macro defines:
//
// - A C-style enum containing all the pass names and a `NoPass` variant.
// - A usize constant with the number of defined passes.
// - A const array of pass descriptions.
// - A public function per pass used to start the timing of that pass.
macro_rules! define_passes {
{ $enum:ident, $num_passes:ident, $descriptions:ident;
$($pass:ident: $desc:expr,)+
} => {
#[allow(non_camel_case_types)]
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
enum $enum { $($pass,)+ NoPass }
const $num_passes: usize = $enum::NoPass as usize;
const $descriptions: [&str; $num_passes] = [ $($desc),+ ];
$(
#[doc=$desc]
pub fn $pass() -> TimingToken {
details::start_pass($enum::$pass)
}
)+
}
}
// Pass definitions.
define_passes!{
Pass, NUM_PASSES, DESCRIPTIONS;
process_file: "Processing test file",
parse_text: "Parsing textual Cretonne IL",
wasm_translate_module: "Translate WASM module",
wasm_translate_function: "Translate WASM function",
verifier: "Verify Cretonne IL",
verify_cssa: "Verify CSSA",
verify_liveness: "Verify live ranges",
verify_locations: "Verify value locations",
verify_flags: "Verify CPU flags",
compile: "Compilation passes",
flowgraph: "Control flow graph",
domtree: "Dominator tree",
loop_analysis: "Loop analysis",
legalize: "Legalization",
gvn: "Global value numbering",
licm: "Loop invariant code motion",
unreachable_code: "Remove unreachable blocks",
regalloc: "Register allocation",
ra_liveness: "RA liveness analysis",
ra_cssa: "RA coalescing CSSA",
ra_spilling: "RA spilling",
ra_reload: "RA reloading",
ra_coloring: "RA coloring",
prologue_epilogue: "Prologue/epilogue insertion",
binemit: "Binary machine code emission",
layout_renumber: "Layout full renumbering",
}
impl Pass {
pub fn idx(self) -> usize {
self as usize
}
}
impl fmt::Display for Pass {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
match DESCRIPTIONS.get(self.idx()) {
Some(s) => f.write_str(s),
None => f.write_str("<no pass>"),
}
}
}
/// Implementation details.
///
/// This whole module can be gated on a `cfg` feature to provide a dummy implementation for
/// performance-sensitive builds or restricted environments. The dummy implementation must provide
/// `TimingToken` and `PassTimings` types and a `take_current` function.
mod details {
use super::{Pass, NUM_PASSES, DESCRIPTIONS};
use std::cell::{Cell, RefCell};
use std::fmt;
use std::mem;
use std::time::{Instant, Duration};
/// A timing token is responsible for timing the currently running pass. Timing starts when it
/// is created and ends when it is dropped.
///
/// Multiple passes can be active at the same time, but they must be started and stopped in a
/// LIFO fashion.
pub struct TimingToken {
/// Start time for this pass.
start: Instant,
// Pass being timed by this token.
pass: Pass,
// The previously active pass which will be restored when this token is dropped.
prev: Pass,
}
/// Accumulated timing information for a single pass.
#[derive(Default)]
struct PassTime {
/// Total time spent running this pas including children.
total: Duration,
/// Time spent running in child passes.
child: Duration,
}
/// Accumulated timing for all passes.
#[derive(Default)]
pub struct PassTimes {
pass: [PassTime; NUM_PASSES],
}
impl fmt::Display for PassTimes {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
writeln!(f, "======== ======== ==================================")?;
writeln!(f, " Total Self Pass")?;
writeln!(f, "-------- -------- ----------------------------------")?;
for (time, desc) in self.pass.iter().zip(&DESCRIPTIONS) {
// Omit passes that haven't run.
if time.total == Duration::default() {
continue;
}
// Write a duration as secs.milis, trailing space.
fn fmtdur(mut dur: Duration, f: &mut fmt::Formatter) -> fmt::Result {
// Round to nearest ms by adding 500us.
dur += Duration::new(0, 500_000);
let ms = dur.subsec_nanos() / 1_000_000;
write!(f, "{:4}.{:03} ", dur.as_secs(), ms)
}
fmtdur(time.total, f)?;
if let Some(s) = time.total.checked_sub(time.child) {
fmtdur(s, f)?;
}
writeln!(f, " {}", desc)?;
}
writeln!(f, "======== ======== ==================================")
}
}
/// Information about passes in a single thread.
thread_local!{
static CURRENT_PASS: Cell<Pass> = Cell::new(Pass::NoPass);
static PASS_TIME: RefCell<PassTimes> = RefCell::new(Default::default());
}
/// Start timing `pass` as a child of the currently running pass, if any.
///
/// This function is called by the publicly exposed pass functions.
pub(super) fn start_pass(pass: Pass) -> TimingToken {
let prev = CURRENT_PASS.with(|p| p.replace(pass));
dbg!("timing: Starting {}, (during {})", pass, prev);
TimingToken {
start: Instant::now(),
pass,
prev,
}
}
/// Dropping a timing token indicated the end of the pass.
impl Drop for TimingToken {
fn drop(&mut self) {
let duration = self.start.elapsed();
dbg!("timing: Ending {}", self.pass);
let old_cur = CURRENT_PASS.with(|p| p.replace(self.prev));
assert_eq!(self.pass, old_cur, "Timing tokens dropped out of order");
PASS_TIME.with(|rc| {
let mut table = rc.borrow_mut();
table.pass[self.pass.idx()].total += duration;
if let Some(parent) = table.pass.get_mut(self.prev.idx()) {
parent.child += duration;
}
})
}
}
/// Take the current accumulated pass timings and reset the timings for the current thread.
pub fn take_current() -> PassTimes {
PASS_TIME.with(|rc| mem::replace(&mut *rc.borrow_mut(), Default::default()))
}
/// Add `timings` to the accumulated timings for the current thread.
pub fn add_to_current(times: PassTimes) {
PASS_TIME.with(|rc| for (a, b) in rc.borrow_mut().pass.iter_mut().zip(
&times.pass,
)
{
a.total += b.total;
a.child += b.child;
})
}
}
#[cfg(test)]
mod test {
use super::*;
#[test]
fn display() {
assert_eq!(Pass::NoPass.to_string(), "<no pass>");
assert_eq!(Pass::regalloc.to_string(), "Register allocation");
}
}