Implement sync_threads using an unaligned barrier. #798
Conversation
maleadt
added
cuda kernels
Codecov Report
@@ Coverage Diff @@
## master #798 +/- ##
=======================================
Coverage 78.46% 78.46%
=======================================
Files 120 120
Lines 7338 7338
=======================================
Hits 5758 5758
Misses 1580 1580 Continue to review full report at Codecov.
|
| """ | ||
| @inline sync_threads() = ccall("llvm.nvvm.barrier0", llvmcall, Cvoid, ()) | ||
| @inline sync_threads(id::Int=0) = ccall("llvm.nvvm.barrier.sync", llvmcall, Cvoid, (Int32,), id) |
There was a problem hiding this comment.
Maybe you can just add sync_threads(id::Int) and keep the old sync_threads()? People can then start using sync_threads(0) to see if it works well in practice.
Also, I don't mind just vendoring this definition and experimenting with it. (If it's only me complaining this thing.)
There was a problem hiding this comment.
You can also do an aligned barrier with a name. So it might be confusing that the named one is unaligned, while the unnamed one is still aligned.
There was a problem hiding this comment.
@tkf Did you end up vendoring this definition then?
There was a problem hiding this comment.
This hasn't come up as a problem and I had no chance to play with it.
maleadt
force-pushed
the
master
branch
24 times, most recently
from
60efe5e to
8704f80
Compare
maleadt
force-pushed
the
master
branch
3 times, most recently
from
a1af3c1 to
46f6109
Compare
maleadt
force-pushed
the
master
branch
15 times, most recently
from
5d585c4 to
c850163
Compare
See #797