tvm
tvm copied to clipboard
apache
Open deep learning compiler stack for cpu, gpu and specialized accelerators
- Introduce `input_axis_separator` in `relax.transform_layout` op for 2D input buffer to allow conversion of 2D buffers to 1D buffers. The conversion from 2D->1D is handled while lowering of `transform_layout` operator....
This PR introduces the Dynamic Gradient Descent (DGD) Search algorithm for accelerating the auto-tuning process of GPU kernels within the Ansor/AutoScheduler framework. The DGD algorithm is designed to explore the...
Add support for executing v75 (Snapdragon 8 gen 3). This PR just adds the support, but to build and execute for v75, the Hexagon SDK used should be 5.4+.
This patch adds `tir.dp4a` as a new TIR built-in operator as a preparation of supporting int8 computation with `dot4I8Packed` in WebGPU backend. issue: https://github.com/apache/tvm/issues/16627
The `ruff format` tool is an alternative to the `black` formatter, with significantly improved performance. This commit updates the `pyproject.toml` to include a configuration for `ruff format`, matched to the...
Prior to this PR, a TIR PrimFunc could return an `int64`, `float64`, or void. Returning any other type, even if supported by TVM's FFI, would raise an exception during `MakePackedAPI`....
Fix a bug in the test_transform_dead_code_elimination.py file about the IR construction.
Add an NNEF frontend to Relax and Relay, as proposed in [RFC #108](https://github.com/apache/tvm-rfcs/pull/108). The Docker image scripts and test scripts have also been extended to accommodate NNEF.
> [Last release v0.16.0](https://github.com/apache/tvm/issues/16857) was proposed at the end of April. and the release day is 27 April, more detail refer [v0.16.0 release schedule](https://github.com/apache/tvm/issues/16857). It has been almost **three months**...
### Actual behavior Segmentation fault (core dumped) ### Environment TVM: 0.17.dev0 OS: Ubuntu20.04 ### Steps to reproduce ``` import tvm from tvm import relax from tvm.script import ir as I...