Philippe Tillet
|
a68061aa96
|
LICENSING: Switched to MIT
|
2017-01-17 20:33:46 -05:00 |
|
Philippe Tillet
|
d52a05db4b
|
Database: Re-tuned Intel Broadwell iGPU for newest driver and double precision
|
2017-01-15 02:56:07 -05:00 |
|
Philippe Tillet
|
aa0322f7d2
|
new Intel json
|
2017-01-09 18:03:39 -05:00 |
|
Philippe Tillet
|
54b5b7523d
|
Core: Added double-precision tuning, tests and benchmarks
|
2016-11-20 22:36:08 -05:00 |
|
Philippe Tillet
|
74e89771ba
|
Database: Added SM-6.0 profile [Contribution from Massimiliano Fatica]
|
2016-11-18 20:15:32 -05:00 |
|
Philippe Tillet
|
48c6592e0c
|
Database: Updated GCN 3 profile
|
2016-10-06 15:24:31 -04:00 |
|
Philippe Tillet
|
06079ab2d6
|
Database: Updated Broadwell profile
|
2016-10-06 14:13:38 -04:00 |
|
Philippe Tillet
|
4e47380127
|
Database: updated SM6.1 profiles
|
2016-10-06 12:36:48 -04:00 |
|
Philippe Tillet
|
fb1205ca7f
|
Database: Updated GCN 3 profile
|
2016-10-06 11:59:42 -04:00 |
|
Philippe Tillet
|
b21024cd37
|
Database: Renamed GCN architectures and added some default profiles
|
2016-10-06 09:51:07 -04:00 |
|
Philippe Tillet
|
625dbf8de7
|
Database: improved SM 6.1 profile
|
2016-10-06 09:11:38 -04:00 |
|
Philippe Tillet
|
1085ea81cc
|
Database: updated SM 3.0 profile
|
2016-10-06 03:10:42 -04:00 |
|
Philippe Tillet
|
ebc2aeba14
|
Tune: More training samples ; better handling of local minima
|
2016-10-06 01:57:42 -04:00 |
|
Philippe Tillet
|
1ece58f5bb
|
Tuner: fixed (benign) bug that caused too many cublas profile creations
|
2016-10-05 19:38:10 -04:00 |
|
Philippe Tillet
|
87cb0ab375
|
Code Quality: Removed useless/buggy fetch_type tuning parameter
|
2016-10-05 19:10:12 -04:00 |
|
Philippe Tillet
|
52fc41461a
|
Elementwise: Bugfix for FETCH_LOCAL_CONTIGUOUS
|
2016-10-05 13:16:58 -04:00 |
|
Philippe Tillet
|
ce9d12ea9d
|
Database: Updated SM6.1 model
|
2016-10-04 23:09:08 -04:00 |
|
Philippe Tillet
|
3293c45e60
|
GEMM: Enabled use of cuBLAS when predicted beneficial
|
2016-10-04 21:17:17 -04:00 |
|
Philippe Tillet
|
294fc96a93
|
Database: Updated Maxwell profile
|
2016-10-03 13:56:58 -04:00 |
|
Philippe Tillet
|
77178d7017
|
GEMM: Better handling of AT=1 and BT=0
|
2016-10-02 17:37:49 -04:00 |
|
Philippe Tillet
|
e1baf85707
|
Code quality: removed obsolete/dead code
|
2016-10-01 19:27:42 -04:00 |
|