tvm
tvm copied to clipboard
apache
Open deep learning compiler stack for cpu, gpu and specialized accelerators
Adds native FP8 type support for CUDA. The e4m3/e5m2 struct types provide explicit type conversions that target hardware native conversion ops. \* Conditionally run Storage and Compute legalization for targets...
Prior to this commit, `int`, `float`, and `bool` arguments from Python were converted to `IntImm`, `FloatImm`, and `Bool`. These are subtypes of `PrimExpr`, and should only be used at compile-time....
Prior to this commit, `FuseOps` and `FuseOpsByPattern` exposed a symbolic variable to the fused function if it was used within the fused function, but wasn't inferable from other parameter shapes....
Internally, `FuseOpsByPattern` makes a mapping from relax variables to the fused group containing that variable. If the input module violates SSA, this map may be ill-formed. While not strictly necessary...
Prior to this commit, using `R.match_cast` as the last binding would produce a segfault, as `var_binding->value` was used instead of `match_cast->value`. In addition, because the last binding of each branch...
Currently, the pytorch wheels available through `pip install` use the pre-C++11 ABI by setting `-DUSE_CXX11_ABI=0` [0]. If TVM were to user the pre-C++11 ABI, this would cause breakages with dynamically-linked...
## Description This pull request aims to enhance model optimization by combining parts of Ansor and AutoTVM. The proposed approach involves the following steps: 1. Execution of Ansor over an...
Prior to this commit, the `MakePackedAPI` pass would output steps in the following order: 1. Check the number of arguments. 2. All `LetStmt` produced by the `ArgBinder` 3. `AssertStmt` for...
Currently, when a default compile target such as `llvm` is specified, it implies `llvm -keys=cpu` which tends to imply x86 related components being used during compilation e.g. the schedules registered...
### Expected behavior I have built oneDNN following this link https://oneapi-src.github.io/oneDNN/dev_guide_build.html. And I modify the config.cmake set(USE_DNNL ON). It should be able to build TVM with DNNL json BYOC feature...
