094890c324
Use shared memory for iter_count and have each CP processed by only one CTA.
...
Slower, but the code is a bit simpler conceptually, and the difference will be
more than accounted for by better scheduling towards the end of the process.
2010-09-07 14:54:50 -04:00
aa065dc25d
Add the first of many microbenchmarks
2010-09-07 12:44:12 -04:00
db72a7d496
Allow register local name rebinding
2010-09-06 16:50:54 -04:00
e03f20392d
Switch from to_inject() to object insertion. One less kludge to deal with.
2010-09-06 16:09:37 -04:00
ada0fe20c7
Random floats (I think)
2010-09-06 14:19:06 -04:00
f3298e0bed
Finally runs again
2010-09-06 11:18:20 -04:00
27e7fd82a3
Time to go have nightmares about this code again (no really)
2010-09-03 00:52:27 -04:00
2c26ff9ab6
* Fix deptrace typos
...
* Add predicate support to DeviceStream fetches, making them even uglier
* Add `store_per_thread` to PTX stdlib
2010-09-03 00:51:23 -04:00
7e7fbda2cc
Start of render module
2010-09-02 17:26:16 -04:00
a23a493d68
Formatter improvements
2010-09-02 16:12:22 -04:00
731c637f80
DataStream. Completely untested. I want to see the bugfixes in the log.
2010-09-02 16:11:44 -04:00
bf79dc7fa0
Adjust softjoin and vec to be easier to format
2010-09-02 15:05:24 -04:00
0a3f9551ad
get_gtid docstring
2010-09-01 23:38:29 -04:00
4a80d946db
Add comment to DSL
2010-09-01 23:34:24 -04:00
32f68ea1d5
Remove some dead code
2010-09-01 22:46:55 -04:00
a3660ec6e4
PTX DSL working, at least well enough to pass MWCRNGTest
2010-09-01 21:09:40 -04:00
5f8c2bbf08
Known broken checkin to show algorias
2010-09-01 13:02:12 -04:00
cceb75396f
Before I rip out tempita and start a DSL
2010-08-30 14:45:44 -04:00
0c78e972b1
Splitting things up a bit
2010-08-28 16:56:05 -04:00
