tvm
tvm copied to clipboard
apache
Open deep learning compiler stack for cpu, gpu and specialized accelerators
If a Relax function contains an `R.match_cast` that defines a symbolic shape, and the value provided to the `R.match_cast` has a known static shape, the `relax.transform.CanoncalizeBindings()` pass can in-line the...
The warp reduction implemented by "shuffle down" primitive takes a mask denoting the active threads within the warp that participate in this shuffle. Previously we compute the mask, while in...
### Actual behavior ``` Traceback (most recent call last): File "/share_container/optfuzz/res/res_ut/res_executions/30_test.py", line 50, in ex = relax.build(mod, target='llvm') ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/software/tvm/python/tvm/relax/vm_build.py", line 335, in build mod = pipeline(mod) ^^^^^^^^^^^^^ File...
This PR refactors the how-to section and add new tutorials of `end-to-end optimization model` cc @tqchen
[Expand](https://onnx.ai/onnx/operators/onnx__Expand.html) dimensions are right alignment. However, the current implementation in onnx_frontend was left-aligned. This PR fixes this bug. @Hzfengsy
Fix crash caused by unhandled return type of `kTVMArgBool`
As per title. Related issue and pr: - #16547 - #17293 - #17297 cc: @Hzfengsy @tqchen
This is a follow-up PR to https://github.com/apache/tvm/pull/16183, which updated the FFI with explicit integer types. As part of that change, many internal functions were updated to accept non-IR types (e.g....
Prior to this commit, the Relax well-formed checker validated arguments provided to Relax functions, but did not validate arguments provided to `R.call_tir`. As a result, incorrect arguments from Relax to...
Prior to this commit, the `shape` argument to `R.reshape` was required to either be an in-line `relax::ShapeExpr`, or a variable that had been bound to a `relax::ShapeExpr` within the current...