Slow parallel C compilation #25203
Description
Nim Version
Nim Compiler Version 2.3.1 [Linux: amd64]
Compiled at 2025-10-03
Copyright (c) 2006-2025 by Andreas Rumpf
git hash: 483389d3999e243d7c61647c55187b0578dc57ee
active boot switches: -d:release
Also on stable and devel releases
Description
The parallel compilation of .c files is slower that it should be when there are many files. (For my project, it is ~30 seconds vs ~4 seconds when executing the same commands outside of nim)
I'm on Ubuntu 24.04, gcc 13.3.0
Here is an example project (depends on chronos and websock):
# test.nim
import std/uri
import chronos
import chronicles
import websock/websock
proc main() {.async.} =
let ws = WebSocket.connect(parseUri("ws://localhost:8080"))
waitFor main()Here is instrumentation:
diff --git a/compiler/extccomp.nim b/compiler/extccomp.nim
index e6e35f462..8232b1609 100644
--- a/compiler/extccomp.nim
+++ b/compiler/extccomp.nim
@@ -993,7 +993,29 @@ proc callCCompiler*(conf: ConfigRef) =
script.add("\n")
if optCompileOnly notin conf.globalOptions:
+ for cmd in cmds:
+ echo cmd
+
+ block:
+ let tStart = getTime()
+ let p = startProcess(
+ "xargs -I{} -P$1 sh -c '{}'" % [$countProcessors()],
+ options={poEvalCommand})
+ let inp = p.inputStream
+ for cmd in cmds:
+ inp.writeLine(cmd)
+ inp.flush()
+ inp.close()
+ let ret = p.waitForExit()
+ let tEnd = getTime()
+ let dt = 1e-3*(tEnd - tStart).inMilliseconds.float
+ stderr.writeLine ">>> compiled in ", dt.formatFloat(ffDecimal, 3)
+
+ let tStart = getTime()
execCmdsInParallel(conf, cmds, prettyCb)
+ let tEnd = getTime()
+ let dt = 1e-3*(tEnd - tStart).inMilliseconds.float
+ stderr.writeLine ">>> compiled in ", dt.formatFloat(ffDecimal, 3)
if optNoLinking notin conf.globalOptions:
# call the linker:
var objfiles = ""
With this, run
nim c -f test.nim > cmds.txt
This will print both measured compiled times and save the commands to a text file.
Current Output
...
>>> compiled in 0.690
CC: ...
...
>>> compiled in 1.616
...
Expected Output
...
>>> compiled in 0.690
CC: ...
...
>>> compiled in 0.690
...
Known Workarounds
No response
Additional Information
Ubuntu 24.04, gcc 13.3.0
I've tried calling execProcesses directly on the saved commands:
import osproc, times, strformat
var cmds: seq[string]
for line in lines("cmds.txt"):
cmds.add(line)
let tStart = getTime()
let ret = execProcesses(
cmds, {poStdErrToStdOut, poUsePath, poParentStreams}, n=countProcessors())
if ret == 0:
let tEnd = getTime()
let dt = 1e-3*((tEnd - tStart).inMilliseconds).float
echo "compiled in {dt:.3f} s".fmt
else:
echo "failed"This gives the same timing as xargs. While the same execProcesses call placed instead of xargs in the compiler takes much more time.
Out of ideas what could be the problem for now.