vvolhejn / thesis

ETH Zürich MSc Thesis: Accelerating Neural Audio Synthesis
Apache License 2.0
17 stars 1 forks source link

Try tuning for TVM #33

Closed vvolhejn closed 2 years ago

vvolhejn commented 2 years ago

There is some issue when tuning the dilated conv models:

Could not find any valid schedule for task Task(func_name=conv2d_NCHWc.x86, args=(('TENSOR', (1, 128, 64000, 1), 'float32'), ('TENSOR', (128, 128, 3, 1), 'float32'), (1, 1), (4, 0, 4, 0), (4, 1), 'NCHW', 'NCHW', 'float32'), kwargs={}, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 128, 64000, 1), 'float32'), ('TENSOR', (128, 128, 3, 1), 'float32'), (1, 1), (4, 0, 4, 0), (4, 1), 'NCHW', 'NCHW', 'float32')). A file containing the errors has been written to /tmp/tvm_tuning_errors_im8wxf8x.log.

Seems related to this issue: https://discuss.tvm.apache.org/t/autoscheduler-failed-to-find-a-valid-schedule/11001

Tuning with dynamic quantization leads to LLVM ERROR: Do not know how to split the result of this operator!

vvolhejn commented 2 years ago

Running into some trouble for inverted bottleneck models as well: https://discuss.tvm.apache.org/t/autotvm-fails-with-could-not-find-any-valid-schedule-for-task-task-func-name-conv2d-nchwc-x86/13024

vvolhejn commented 2 years ago

Decided not to do this due to the issues mentioned above.