We should see if LTO improves the performance of OpenROAD

[QuantamHD]

LTO stands for link time optimization, and allows clang/gcc to optimize across linking boundaries.

We can enable it by default by setting set(CMAKE_INTERPROCEDURAL_OPTIMIZATION TRUE) in the root CMAKE file.

https://stackoverflow.com/questions/31355692/how-do-i-enable-link-time-optimization-lto-with-cmake

[maliberty]

@QuantamHD do you want to try it?

[cbalint13]

@QuantamHD @maliberty

OpenROAD +LTO means:

• 636,005,720 vs 830,627,144 unstripped binary.
• 44,501,056 vs 50,634,568 stripped binary.
• 3m56.108s vs 3m57.196s regression runtime.

LTO is more about size optimisation (speed would second by a chance) IMO is safe to enable LTO it gains some significant amount towards smaller size.

---

Setup conditions:

• OpenROAD git hash: a1cd122fe9bf2e6cb7b2b05be43c50dc5599d8f2

• GCC version:

  $ gcc -v
  Supported LTO compression algorithms: zlib zstd
  gcc version 12.0.1 20220401 (Red Hat 12.0.1-0) (GCC)

• OpenROAD having external OpenSTA & abc (booth had LTO as per Fedora default):

  $ readelf -a /usr/bin/openroads | grep NEED
  0x0000000000000001 (NEEDED)             Shared library: [libOpenSTA.so.2]
  0x0000000000000001 (NEEDED)             Shared library: [libabc.so.0]
  0x0000000000000001 (NEEDED)             Shared library: [libtcl8.6.so]
  0x0000000000000001 (NEEDED)             Shared library: [libtclreadline-2.1.0.so]
  0x0000000000000001 (NEEDED)             Shared library: [libX11.so.6]
  0x0000000000000001 (NEEDED)             Shared library: [libjpeg.so.62]
  0x0000000000000001 (NEEDED)             Shared library: [libgomp.so.1]
  0x0000000000000001 (NEEDED)             Shared library: [libboost_serialization.so.1.76.0]
  0x0000000000000001 (NEEDED)             Shared library: [libQt5Widgets.so.5]
  0x0000000000000001 (NEEDED)             Shared library: [libQt5Gui.so.5]
  0x0000000000000001 (NEEDED)             Shared library: [libQt5Core.so.5]
  0x0000000000000001 (NEEDED)             Shared library: [libspdlog.so.1]
  0x0000000000000001 (NEEDED)             Shared library: [libfmt.so.8]
  0x0000000000000001 (NEEDED)             Shared library: [libz.so.1]
  0x0000000000000001 (NEEDED)             Shared library: [libpython3.10.so.1.0]
  0x0000000000000001 (NEEDED)             Shared library: [libstdc++.so.6]
  0x0000000000000001 (NEEDED)             Shared library: [libm.so.6]
  0x0000000000000001 (NEEDED)             Shared library: [libgcc_s.so.1]
  0x0000000000000001 (NEEDED)             Shared library: [libc.so.6]

---

LTO on:

• Extra flags:

  $ rpm --eval %{_lto_cflags}
  -flto=auto -ffat-lto-objects

• Size unstripped:

  $ ls -l build/src/openroad
  -rwxr-xr-x 1 cbalint cbalint 636005720 Apr 10 17:08 build/src/openroad

• Size stripped:

  $ ls -l /usr/bin/openroad
  -rwxr-xr-x 1 cbalint cbalint 44501056 Apr 10 17:09 /usr/bin/openroad

• Regression timing:

  Passed all tool regressions.
  real    3m56.108s
  user    3m38.084s
  sys 0m14.937s

LTO off:

• Size unstripped:

  $ ls -l build/src/openroad
  -rwxr-xr-x 1 cbalint cbalint 830627144 Apr 10 17:20 build/src/openroad

• Size stripped:

  ls -l /usr/bin/openroad
  -rwxr-xr-x 1 cbalint cbalint 50634568 Apr 10 17:21 /usr/bin/openroad

• Regression timing:

  Passed all tool regressions.
  real    3m57.196s
  user    3m42.273s
  sys 0m14.460s

[QuantamHD]

LTO was broken because of the QT components. We would need some additional work to selectively disable it around the GUI.

[QuantamHD]

Auto-FDO would also be worth doing.