tvm
tvm copied to clipboard
apache
Open deep learning compiler stack for cpu, gpu and specialized accelerators
Automatic unrolling in LLVM was causing significant issues with compile time and binary size of a bert int8 model. This patch disables unrolling in LLVM for the hexagon backend. While...
See https://github.com/apache/tvm/actions/runs/3097181204/jobs/5013784930. ``` Traceback (most recent call last): File "ci/scripts/github_pr_comment.py", line 132, in skipped_tests = get_skipped_tests_comment(pr_data, github=github) File "/home/runner/work/tvm/tvm/ci/scripts/github_skipped_tests_comment.py", line 164, in get_skipped_tests_comment common_main_build = get_main_jenkins_build_number(github, common_commit_sha) File "/home/runner/work/tvm/tvm/ci/scripts/github_skipped_tests_comment.py", line...
After #12720 the RewriteLayout postprocessor seems to fail during tuning. An example to reproduce is here https://gist.github.com/zxybazh/6bff29ae4e7cb273d57bb30599790008. And the failing message looks like: ``` [11:43:50] /home/zxybazh/tvm-tensorir/src/meta_schedule/search_strategy/../utils.h:289: Warning: ThreadedTraceApply::Apply failed with...
For a while, I've intended to fix my depthwise_conv2d schedule so that its unique weight repacking scheme happens at compile time instead of during inference. While working on this though,...
Thanks for participating in the TVM community! We use https://discuss.tvm.ai for any general usage questions and discussions. The issue tracker is used for actionable items such as feature proposals discussion,...
New docker for Adreno that inherits gpu docker and amends android-sdk. For CLML, we need to specify ADRENO_OPENCL via environment variable. CLML SDK can be downlaoded from Qualcomm Developer Network...
Currently, cache read/write requires to be stage pipeline, but it is unnecessary theoretically. When there is WAR, the target of cache_read could be specified by the consumer_blocks parameter. This also...
Building on https://github.com/apache/tvm/pull/12845, this PR adds an initial support for template-free auto tuning on Hexagon. Test cases demonstrate: * Auto-scheduler style, template free tuning for fp16 conv2d in NHWC layout....
Potentially, there is an issue with implementation of PyTorch `aten::fill_()` operator in the PyTorch frontend. I have found this bug when I was trying to run [gallery/how_to/deploy_models/deploy_object_detection_pytorch.py](https://github.com/apache/tvm/blob/main/gallery/how_to/deploy_models/deploy_object_detection_pytorch.py) example with Torch...
This adds a `--tasks` flag to the `tvmc tune` command to filter the lists of tasks to be tuned. See examples below. ## Motivation - As auto-tuning can be quite...
