#4204 closed enhancement (wontfix)
Xorg Intel Driver
| Reported by: | jon lee | Owned by: | |
|---|---|---|---|
| Priority: | normal | Milestone: | |
| Component: | BOOK | Version: | SVN |
| Severity: | normal | Keywords: | Intel Xorg |
| Cc: |
Description
I noticed a configure check for clang in the Intel Xorg driver, found here: http://www.linuxfromscratch.org/blfs/view/svn/x/x7driver.html
Since clang has to be used to build the mesa radeon drivers, (and reportedly improves performance there), I attempted to use it to build the Intel driver.
Using: export CC=clang export CXX=clang export CFLAGS="-std=gnu89 -g"
before the ./configure... this will improve compile time greatly and the resulting driver performs much better as well.
Change History (6)
comment:1 by , 11 years ago
comment:2 by , 10 years ago
I don't have an Intel card, so I can't check this, but it appears that we should add llvm as an optional dependency and put some notes into the Command Explanations section. That should do it for this ticket.
comment:3 by , 10 years ago
I do have intel hardware, but I'm still waiting for performance metrics. Those of us using KMS (that should include everyone using the intel driver) know that frames per second is synchronized to the vertical refresh rate, so I don't have any idea *how* a runtime performance improvement is being measured. I'm not ruling it out, just reserving judgement.
Much, if not most, of the book can probably be built with clang. At least one package had a problem with it in the past few months, but so far I don't see any reason to add a comment about using clang for this particular package.
Equally, I don't think that any benefit from faster compilation, if it exists, will be material. On my 7.4 intel machine, configure for the driver takes 4.699s, make -j1 takes 47.003s, and a DESTDIR install 0.306s. Call it 52 seconds.
Until we see some hard evidence, this sounds suspiciously like adding go-faster stripes to a car.
follow-up: 5 comment:4 by , 10 years ago
If the intel driver developers didn't bother to search for clang at configure time (see icu for example) then I'd rather not mention anything at all.
comment:5 by , 10 years ago
| Resolution: | → wontfix |
|---|---|
| Status: | new → closed |
Replying to izivkov:
If the intel driver developers didn't bother to search for clang at configure time (see icu for example) then I'd rather not mention anything at all.
Good Point. Closing.
First, clang does not have to be used for the mesa radeon drivers - llvm without clang builds then fine.
Do you have any metrics for the performance improvement ?
On my intel box, I normally use make -j4 for the driver and the most recent build took less than 22 seconds. Even if that becomes 88 seconds with -j1, and maybe two minutes on a really slow box, will you recover the extra time from adding clang into your llvm build ?
To me, this looks like the sort of thing you could put in the wiki.