c6ed99a | aekul | 30 July 2021, 01:02:12 UTC | Fix validation set in retrain_cost_model.cpp; fix limit flag | 30 July 2021, 02:15:36 UTC |
31f7e8b | aekul | 16 April 2021, 16:55:56 UTC | Enable benchmark queue as default | 30 July 2021, 02:15:14 UTC |
e6853fd | aekul | 05 April 2021, 20:39:52 UTC | Fix missing outer gpu_blocks loop | 30 July 2021, 02:07:57 UTC |
8ee4487 | aekul | 21 August 2020, 14:00:56 UTC | Demosaic | 30 July 2021, 01:20:40 UTC |
0c67f41 | aekul | 20 August 2020, 20:48:04 UTC | Update Makefiles with weights | 30 July 2021, 01:20:37 UTC |
1b73beb | aekul | 12 July 2021, 15:47:52 UTC | Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 12 July 2021, 15:47:52 UTC |
1d86dc9 | aekul | 15 June 2021, 15:50:19 UTC | Update depthwise_separable_conv generator | 15 June 2021, 15:50:19 UTC |
bd31193 | aekul | 01 December 2020, 03:56:06 UTC | Add app_predictions call to generate_autotune_results.sh | 01 December 2020, 03:56:06 UTC |
aa04ea5 | aekul | 01 December 2020, 02:40:55 UTC | Add prediction scripts | 01 December 2020, 02:40:55 UTC |
583d66e | Andrew Adams | 21 August 2020, 18:23:33 UTC | Inline boundary condition | 21 August 2020, 18:23:33 UTC |
b055b44 | Andrew Adams | 21 August 2020, 18:07:26 UTC | depthwise schedule | 21 August 2020, 18:07:26 UTC |
387c556 | Andrew Adams | 21 August 2020, 14:30:39 UTC | Fix harris manual schedule | 21 August 2020, 14:30:39 UTC |
e945c55 | Andrew Adams | 21 August 2020, 14:00:52 UTC | Manual ahd schedule. Improve basic demosaic schedule | 21 August 2020, 14:00:52 UTC |
dc69509 | Andrew Adams | 21 August 2020, 13:48:51 UTC | Restore input estimates | 21 August 2020, 13:48:51 UTC |
d0c37f0 | Andrew Adams | 21 August 2020, 13:35:16 UTC | Reformat ahd | 21 August 2020, 13:35:16 UTC |
280d9f0 | Andrew Adams | 21 August 2020, 13:31:26 UTC | Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 21 August 2020, 13:31:26 UTC |
cc80929 | Andrew Adams | 21 August 2020, 13:26:36 UTC | manual schedule for basic demosaic | 21 August 2020, 13:26:36 UTC |
802ee95 | Karima Ma | 21 August 2020, 04:14:24 UTC | fixed bugs in relu layer definition and missing relu tensor declarations | 21 August 2020, 04:14:24 UTC |
c222633 | Karima Ma | 21 August 2020, 03:38:25 UTC | fixed missing relus in multires demosaic | 21 August 2020, 03:38:25 UTC |
4c6ba7a | Karima Ma | 21 August 2020, 01:52:04 UTC | explicitly only access green prediction at channel 0 | 21 August 2020, 01:52:04 UTC |
c081923 | aekul | 20 August 2020, 21:08:31 UTC | Replace hist in app list | 20 August 2020, 21:08:31 UTC |
ea555b7 | aekul | 20 August 2020, 21:05:46 UTC | Make batch size 80 | 20 August 2020, 21:05:46 UTC |
b4b2e28 | aekul | 20 August 2020, 21:05:15 UTC | Fix retrain after batch | 20 August 2020, 21:05:15 UTC |
67e8a93 | aekul | 20 August 2020, 20:56:17 UTC | Fix limit flag | 20 August 2020, 20:56:17 UTC |
142d833 | aekul | 20 August 2020, 17:02:19 UTC | Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 20 August 2020, 17:02:19 UTC |
bc372de | aekul | 20 August 2020, 16:28:36 UTC | Wait for compile jobs | 20 August 2020, 16:28:36 UTC |
ca75a10 | Tzu-Mao Li | 20 August 2020, 15:17:34 UTC | preallocate GPU memory for onnx Halide time measurement | 20 August 2020, 15:17:34 UTC |
79f08ac | Andrew Adams | 20 August 2020, 15:02:45 UTC | Add hold-one-out weights from 500 epochs of training Also change the cost model activation function to have a slight leak. Also change the retraining program to do full epochs with 64-sample batches in random order, instead of one batch of size up to 1024 per pipeline. | 20 August 2020, 15:02:45 UTC |
9fd2bfc | Tzu-Mao Li | 20 August 2020, 13:55:34 UTC | fix timing measurement | 20 August 2020, 13:55:34 UTC |
e143c4d | Tzu-Mao Li | 20 August 2020, 02:12:03 UTC | Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 20 August 2020, 02:12:03 UTC |
a5c6370 | Tzu-Mao Li | 20 August 2020, 02:11:34 UTC | fix onnx model benchmarking on satori | 20 August 2020, 02:11:34 UTC |
45b1f4f | aekul | 20 August 2020, 00:35:12 UTC | Don't wait for compilation between batches | 20 August 2020, 00:35:12 UTC |
a032a27 | aekul | 19 August 2020, 23:56:57 UTC | Add NUM_BATCHES back | 19 August 2020, 23:56:57 UTC |
9c139bf | aekul | 19 August 2020, 23:43:43 UTC | Tidy up generate_autotune_results | 19 August 2020, 23:43:43 UTC |
ea47078 | aekul | 19 August 2020, 23:30:43 UTC | Random batch IDs for apps | 19 August 2020, 23:30:43 UTC |
d811abe | Karima Ma | 19 August 2020, 18:11:19 UTC | fixed bugs in building demosaicing apps | 19 August 2020, 18:11:19 UTC |
89ab06b | Tzu-Mao Li | 19 August 2020, 15:31:10 UTC | mobilenet testing | 19 August 2020, 15:31:10 UTC |
f701f98 | aekul | 18 August 2020, 22:46:39 UTC | Add limit to retrain_cost_model | 18 August 2020, 22:46:39 UTC |
c49eac7 | aekul | 18 August 2020, 21:49:21 UTC | Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 18 August 2020, 21:49:21 UTC |
7b13eab | aekul | 18 August 2020, 21:49:07 UTC | Add sample load time to retrain_cost_model.cpp | 18 August 2020, 21:49:07 UTC |
2a2d98e | Karima Ma | 18 August 2020, 02:01:27 UTC | Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 18 August 2020, 02:01:27 UTC |
93d23e4 | Karima Ma | 18 August 2020, 02:01:17 UTC | fixed bugs in demosaic apps | 18 August 2020, 02:01:17 UTC |
f0a5cfa | Tzu-Mao Li | 18 August 2020, 01:53:36 UTC | Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 18 August 2020, 01:53:36 UTC |
fc09851 | Tzu-Mao Li | 18 August 2020, 01:53:15 UTC | fix benchmark models and add tensorflow comparison | 18 August 2020, 01:53:15 UTC |
cdfd950 | Andrew Adams | 17 August 2020, 23:40:40 UTC | Some progress on getting multires working | 17 August 2020, 23:40:40 UTC |
062777d | Andrew Adams | 17 August 2020, 23:11:31 UTC | Reschedule train_cost_model. Use 8 threads when retraining. | 17 August 2020, 23:11:31 UTC |
0856036 | aekul | 17 August 2020, 19:23:46 UTC | Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 17 August 2020, 19:23:46 UTC |
cb4591e | aekul | 17 August 2020, 19:23:44 UTC | Add timing information to retrain_cost_model.cpp | 17 August 2020, 19:23:44 UTC |
782a166 | Jonathan Ragan-Kelley | 17 August 2020, 16:19:46 UTC | Update random pipeline data slurm script * Generate more samples * Shut off core dump files! | 17 August 2020, 16:20:18 UTC |
5a509bd | Karima Ma | 17 August 2020, 16:18:53 UTC | adding process | 17 August 2020, 16:18:53 UTC |
7ed4f09 | Karima Ma | 17 August 2020, 16:09:26 UTC | :Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 17 August 2020, 16:09:26 UTC |
d3c90f7 | Karima Ma | 17 August 2020, 16:09:09 UTC | adding demosaicing apps | 17 August 2020, 16:09:09 UTC |
88a8269 | aekul | 17 August 2020, 14:26:13 UTC | Add epoch completed message | 17 August 2020, 14:26:13 UTC |
461b64a | Tzu-Mao Li | 17 August 2020, 13:56:45 UTC | modify burst camera pipe (didn't work) | 17 August 2020, 13:56:45 UTC |
d77ac75 | aekul | 17 August 2020, 13:42:53 UTC | Fix stats for mobilenet | 17 August 2020, 13:42:53 UTC |
e64ec2b | aekul | 16 August 2020, 13:48:56 UTC | Always consider inline options for partial schedule nodes | 17 August 2020, 02:40:14 UTC |
2bc7025 | aekul | 15 August 2020, 14:15:55 UTC | Add partial schedule support | 17 August 2020, 02:40:14 UTC |
035060e | aekul | 14 August 2020, 16:57:13 UTC | Use <= when checking amount of shared mem | 17 August 2020, 02:40:14 UTC |
5d0766f | aekul | 14 August 2020, 16:35:43 UTC | Skip candidate compute locations with allocations that are too large | 17 August 2020, 02:40:13 UTC |
b31d52e | aekul | 14 August 2020, 15:10:04 UTC | Skip candidate compute locations that use too much shared memory | 17 August 2020, 02:40:13 UTC |
e0e43f0 | aekul | 14 August 2020, 14:22:58 UTC | Add state tests | 17 August 2020, 02:40:13 UTC |
61bb1c1 | aekul | 14 August 2020, 14:21:53 UTC | For inlined funcs, consider all its consumers when finding the deepest common ancestor | 17 August 2020, 02:40:13 UTC |
4620c4f | aekul | 14 August 2020, 14:20:46 UTC | Use active threads when computing warp utilization | 17 August 2020, 02:40:13 UTC |
b5ba27c | Andrew Adams | 17 August 2020, 00:27:24 UTC | Delete dead TODO | 17 August 2020, 00:27:24 UTC |
e989614 | Andrew Adams | 17 August 2020, 00:26:09 UTC | Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 17 August 2020, 00:26:09 UTC |
fa553cd | Andrew Adams | 17 August 2020, 00:26:02 UTC | Fast random initialization using xorshift | 17 August 2020, 00:26:02 UTC |
b60615e | Tzu-Mao Li | 16 August 2020, 22:36:12 UTC | fix run all gradient autoschedule | 16 August 2020, 22:36:12 UTC |
dc9795a | Tzu-Mao Li | 16 August 2020, 22:25:14 UTC | Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 16 August 2020, 22:25:14 UTC |
165c23b | Tzu-Mao Li | 16 August 2020, 22:24:38 UTC | add no gradient autoscheduler flag | 16 August 2020, 22:24:38 UTC |
edaa15f | aekul | 16 August 2020, 20:56:09 UTC | Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 16 August 2020, 20:56:09 UTC |
d6c8a84 | aekul | 16 August 2020, 20:55:59 UTC | Use RunGenMain.o | 16 August 2020, 20:55:59 UTC |
c079842 | Andrew Adams | 16 August 2020, 20:48:52 UTC | Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 16 August 2020, 20:48:52 UTC |
1f8296e | Andrew Adams | 16 August 2020, 20:48:31 UTC | Reschedule BGU This makes it less sensitive to atomic add vs cas loop flakiness | 16 August 2020, 20:48:31 UTC |
e118cbc | aekul | 16 August 2020, 19:38:16 UTC | Add missing .mat files | 16 August 2020, 19:38:16 UTC |
310b34d | aekul | 16 August 2020, 19:26:33 UTC | Add -O3 when compiling RunGen | 16 August 2020, 19:26:33 UTC |
c693325 | aekul | 16 August 2020, 19:13:58 UTC | Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 16 August 2020, 19:13:58 UTC |
a5cbad4 | aekul | 16 August 2020, 19:13:51 UTC | Add manual script | 16 August 2020, 19:13:51 UTC |
42791b1 | aekul | 16 August 2020, 18:22:45 UTC | Add missing copy_to_host calls | 16 August 2020, 18:22:45 UTC |
2629b01 | Tzu-Mao Li | 16 August 2020, 18:11:51 UTC | remove unused script | 16 August 2020, 18:11:51 UTC |
1c6f3e7 | Tzu-Mao Li | 16 August 2020, 18:09:03 UTC | mobilenet script | 16 August 2020, 18:09:03 UTC |
10d1f9d | aekul | 16 August 2020, 17:28:29 UTC | Add missing image files | 16 August 2020, 17:28:29 UTC |
1fbb949 | aekul | 16 August 2020, 16:56:24 UTC | Add missing header and copy_to_host to iir_blur | 16 August 2020, 16:56:24 UTC |
2b78dd7 | Andrew Adams | 16 August 2020, 16:07:39 UTC | Set max registers in ptxas command | 16 August 2020, 16:07:39 UTC |
6c822b7 | aekul | 15 August 2020, 15:23:29 UTC | Check if best_schedule file is missing | 15 August 2020, 15:23:29 UTC |
e7f2176 | aekul | 15 August 2020, 14:54:45 UTC | Add 'loading samples' message | 15 August 2020, 14:55:04 UTC |
f5d27c5 | aekul | 15 August 2020, 14:36:51 UTC | Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 15 August 2020, 14:36:51 UTC |
b72941c | aekul | 15 August 2020, 14:36:08 UTC | Remove files after generating data | 15 August 2020, 14:36:08 UTC |
a7a633f | Jonathan Ragan-Kelley | 15 August 2020, 06:10:18 UTC | Make random_pipeline zero-initialize buffers for speed during data generation | 15 August 2020, 06:10:18 UTC |
0721482 | Jonathan Ragan-Kelley | 15 August 2020, 03:39:52 UTC | Adding autotuning helpers: - Slurm script for random pipeline data gathering on Satori - Slurm one-liner to launch interactive session on Satori | 15 August 2020, 03:39:52 UTC |
bdb4cf4 | aekul | 15 August 2020, 02:19:36 UTC | Add Func names to lens_blur_generator.cpp | 15 August 2020, 02:19:36 UTC |
1436f6b | aekul | 15 August 2020, 01:26:13 UTC | Fix benchmark queue timeout | 15 August 2020, 01:26:13 UTC |
5b72a8b | Jonathan Ragan-Kelley | 15 August 2020, 01:15:35 UTC | Update generate_autotune_results to compute capability 7.0 | 15 August 2020, 01:15:35 UTC |
394e23c | Andrew Adams | 15 August 2020, 00:02:39 UTC | Merge branch 'standalone_autoscheduler_gpu' of https://github.com/halide/Halide into standalone_autoscheduler_gpu | 15 August 2020, 00:02:39 UTC |
683d4b5 | Andrew Adams | 15 August 2020, 00:02:29 UTC | Add more cuda generations. Compile to SASS if possible | 15 August 2020, 00:02:29 UTC |
2c75927 | aekul | 14 August 2020, 23:07:13 UTC | Order benchmark files by time | 14 August 2020, 23:07:13 UTC |
476f971 | Andrew Adams | 14 August 2020, 22:29:07 UTC | Add copy-to-host to lens blur | 14 August 2020, 22:29:07 UTC |
d669c68 | aekul | 14 August 2020, 21:38:44 UTC | Add multi-batch support to autotune_loop | 14 August 2020, 21:39:33 UTC |
31f0eb5 | aekul | 14 August 2020, 13:54:27 UTC | Remove scatter plot data point limit | 14 August 2020, 13:54:27 UTC |
80526b7 | aekul | 14 August 2020, 00:45:07 UTC | Fix date in generate_autotune_results.sh | 14 August 2020, 00:45:07 UTC |
46e3933 | aekul | 13 August 2020, 19:48:16 UTC | Don't print name if node is null | 13 August 2020, 19:48:16 UTC |