Commit Graph

263 Commits

Author SHA1 Message Date
Philippe Tillet
397d76156b progress on re-association 2019-07-23 17:21:24 -07:00
Philippe Tillet
38b3771c26 some reassociation 2019-07-23 14:43:18 -07:00
Philippe Tillet
c448876178 better benchmarking 2019-07-22 19:26:12 -07:00
Philippe Tillet
ead368d1ed [general] a bunch of fixes in anticipation of proper triton vs cudnn
benchmarks

* DNN: Added partial auto-tuning mode and skeleton for heuristics
* Examples: Moduralized benchmarking and now evaluating ResNet-18 shapes
2019-07-21 20:17:56 -07:00
Philippe Tillet
b1d81a5802 more work on heuristics 2019-07-21 18:11:54 -07:00
Philippe Tillet
484e3871cf [dnn/shift] added base pointer for a, b 2019-07-20 23:00:27 -07:00
Philippe Tillet
d159455f7b [codegen/alignment_info] better alignment information 2019-07-20 21:44:18 -07:00
Philippe Tillet
28c250216c [dnn/gemm] added some bounds checking 2019-07-19 21:32:55 -07:00
Philippe Tillet
5215fb0424 [codegen] some more optimizations 2019-07-19 20:29:03 -07:00
Philippe Tillet
71594da66f [dnn/gemm]: fixed leading dimension in transposed variants 2019-07-18 16:35:48 -07:00
Philippe Tillet
f0d8306437 [codegen/alignment_info] better handling of constants 2019-07-18 16:12:06 -07:00
Philippe Tillet
86f70f8224 [codegen/selection] performance fix-up when A is transposed for hmma 2019-07-17 21:46:23 -07:00
Philippe Tillet
2f0817b2cd [codegen/selection] tensor cores now used for transposed layotus 2019-07-17 17:20:38 -07:00
Philippe Tillet
bfa39b8992 preparing the field for tensor cores transposes 2019-07-17 13:20:33 -07:00
Philippe Tillet
d2e116d057 testing GEMM 2019-07-17 12:38:30 -07:00
Philippe Tillet
791c91ee63 [dnn/shift] bugfix in static shape division 2019-07-17 11:39:17 -07:00
Philippe Tillet
a55b098e88 [dnn/shift] now using constant divisions 2019-07-16 21:05:21 -07:00
Philippe Tillet
07c964919c [dnn/shift] now strictly only shifting the interior 2019-07-16 20:18:48 -07:00
Philippe Tillet
ec24e1e7df trying to remove interior logic 2019-07-16 18:47:50 -07:00
Philippe Tillet
5f6dd23fc2 [dnn/dot] reverted back to peak tensorcores performance 2019-07-16 16:14:58 -07:00
Philippe Tillet
164d85077f more stuff 2019-07-16 15:03:53 -07:00
Philippe Tillet
28959fe165 [runtime/jit] made auto-tuning silent 2019-07-16 14:41:38 -07:00
Philippe Tillet
7d1797cd32 ugh 2019-07-16 12:59:27 -07:00
Philippe Tillet
f50d7a420a [runtime/jit] fixed bug in multi-threaded auto-tuning 2019-07-15 21:16:50 -07:00
Philippe Tillet
aa8bcf6bde [dnn/shift] added split-k for shift-conv 2019-07-15 21:03:58 -07:00
Philippe Tillet
434f65737f [runtime] put jit::launch_info in another file 2019-07-15 12:35:53 -07:00
Philippe Tillet
3c128fc2e2 [jit/autotune] added support for multi-threaded auto-tuning 2019-07-14 22:31:30 -07:00
Philippe Tillet
3e7a3ed67a [dnn/shift]: added support for fp16 2019-07-13 21:05:34 -07:00
Philippe Tillet
fe42cb7142 [dnn/shift] optimizations for NCHW layout 2019-07-12 20:22:32 -07:00
Philippe Tillet
54617b4e51 some cleaning 2019-07-12 20:10:15 -07:00
Philippe Tillet
7512c7ebed some cleaning 2019-07-12 20:03:05 -07:00
Philippe Tillet
c1c7062914 blabla 2019-07-12 17:42:29 -07:00
Philippe Tillet
f36a646ffc [dnn/shift-conv] added and tested NCHW layout 2019-07-11 21:00:33 -07:00
Philippe Tillet
fe8caf12f0 [dnn/conv]: skeleton for NCHW layout 2019-07-11 20:34:38 -07:00
Philippe Tillet
207e021973 [codegen/shift] substantial cleaning of triton-c shift-conv code 2019-07-11 20:11:23 -07:00
Philippe Tillet
75cf2df110 [dnn/shift] many bugfixes in strided shift-conv 2019-07-10 19:49:31 -07:00
Philippe Tillet
4ca83f1935 ugh bug in shift-conv striding 2019-07-10 17:00:22 -07:00
Philippe Tillet
f665c742f9 testing a simple shiftnet 2019-07-10 13:33:08 -07:00
Philippe Tillet
3b89bc8463 [examples/python/pytorch] added skeleton of wrapper for shift-conv and batch-norm 2019-07-09 21:54:37 -07:00
Philippe Tillet
63b249c1d6 [examples/python/pytorch] added batchnorm cpp extension 2019-07-09 20:59:04 -07:00
Philippe Tillet
b7986baffa [dnn]: Now implementing all existing DNN routines using common base template and auto-tuner 2019-07-09 19:52:55 -07:00
Philippe Tillet
88675fa01a [dnn] added base template class for mutualized auto-tuning 2019-07-09 16:09:34 -07:00
Philippe Tillet
066ae338f1 [dnn/shift]: added stride to shift 2019-07-09 14:08:51 -07:00
Philippe Tillet
cc41604784 [codegen/batchnorm] forward and backward now seemingly working 2019-07-09 13:03:16 -07:00
Philippe Tillet
f74dcb7e30 [dnn/batchnorm]: added some more code in Triton-C batchnorm implementations 2019-07-08 20:18:20 -07:00
Philippe Tillet
fa3270dcf2 [codegen/selection] bugfix in code generation for reduction instructions 2019-07-08 18:53:37 -07:00
Philippe Tillet
f9db0449b7 [dnn] Adding batchnorm 2019-07-08 18:44:37 -07:00
Philippe Tillet
b0cf3143c5 [dnn/shift] bugfix in wgrad 2019-07-06 11:27:49 -07:00
Philippe Tillet
3e49dbe6ab [dnn/shift] fixed in leading dimensions for shift-conv operation 2019-07-05 17:17:22 -07:00
Philippe Tillet
c666f71fd6 fixed bug 2019-07-05 15:07:20 -07:00