[Issue 7157] Optimiser is O(n^2) w.r.t. function length

https://issues.dlang.org/show_bug.cgi?id=7157

Martin Nowak <code@dawg.eu> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |code@dawg.eu

--- Comment #4 from Martin Nowak <code@dawg.eu> ---
I'm getting excessive compile times for a few phobos modules in release
unittests.
Most notably std.regex.internal.tests, and std.datetime.
Can we try to do the static foreach delegate trick again?

time ../dmd/src/dmd -conf= -I../druntime/import  -w -dip25 -m64  -O -release
-unittest -c -ofgenerated/osx/release/64/unittest/std/datetime.o
-deps=generated/osx/release/64/unittest/std/datetime.deps.tmp std/datetime.d
      180.96 real       132.19 user        48.76 sys
time ../dmd/src/dmd -conf= -I../druntime/import  -w -dip25 -m64  -O -release
-unittest -c -ofgenerated/osx/release/64/unittest/std/random.o
-deps=generated/osx/release/64/unittest/std/random.deps.tmp std/random.d
       21.23 real        15.68 user         5.53 sys
time ../dmd/src/dmd -conf= -I../druntime/import  -w -dip25 -m64  -O -release
-unittest -c -ofgenerated/osx/release/64/unittest/std/regex/internal/tests.o
-deps=generated/osx/release/64/unittest/std/regex/internal/tests.deps.tmp
std/regex/internal/tests.d
      250.31 real       184.75 user        65.55 sys

--

https://issues.dlang.org/show_bug.cgi?id=7157 --- Comment #5 from Martin Nowak <code@dawg.eu> --- Most CPU time is spent in `updaterd` and `accumaecpx` if that helps. https://github.com/D-Programming-Language/dmd/blob/82b031c22f18334a7edeff437260b6802070a9e3/src/backend/gloop.c#L1263 https://github.com/D-Programming-Language/dmd/blob/e43136308f7eb035d046a99a46ffbdaf880413b7/src/backend/gflow.c#L907 --

https://issues.dlang.org/show_bug.cgi?id=7157 Jonathan M Davis <issues.dlang@jmdavisProg.com> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |issues.dlang@jmdavisProg.co | |m --- Comment #6 from Jonathan M Davis <issues.dlang@jmdavisProg.com> --- (In reply to Martin Nowak from comment #4) > Can we try to do the static foreach delegate trick again? Which trick are you referring to? --

https://issues.dlang.org/show_bug.cgi?id=7157 --- Comment #7 from Martin Nowak <code@dawg.eu> --- (In reply to Jonathan M Davis from comment #6) > Which trick are you referring to? https://github.com/D-Programming-Language/phobos/pull/2767 --

https://issues.dlang.org/show_bug.cgi?id=7157 --- Comment #8 from Martin Nowak <code@dawg.eu> --- Another case of this. https://github.com/etcimon/botan/issues/8 It took dmd almost 13 minutes to compile botan w/ release optimizations on heroku (and about 4 minutes on my local machine). --

https://issues.dlang.org/show_bug.cgi?id=7157 --- Comment #9 from Martin Nowak <code@dawg.eu> --- Could we fix this for now by letting dmd backoff from very big functions? Or if someone understands accumaecpx well enough, maybe the optimization can be windowed to 1000 expressions/statements or so. --

https://issues.dlang.org/show_bug.cgi?id=7157 Daniel Kozak <kozzi11@gmail.com> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |chalucha@gmail.com --- Comment #10 from Daniel Kozak <kozzi11@gmail.com> --- *** Issue 14939 has been marked as a duplicate of this issue. *** --

https://issues.dlang.org/show_bug.cgi?id=7157 Andrei Alexandrescu <andrei@erdani.com> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |andrei@erdani.com --- Comment #11 from Andrei Alexandrescu <andrei@erdani.com> --- I just realized that a precise GC using introspection would use exactly this facility. --

https://issues.dlang.org/show_bug.cgi?id=7157 Walter Bright <bugzilla@digitalmars.com> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |bugzilla@digitalmars.com --- Comment #12 from Walter Bright <bugzilla@digitalmars.com> --- There is some quadratic behavior in the function bodies of `updaterd` and `accumaecpx` (look for the loops). They can both be fixed. --

Forums