Cookiee235

I came across a similar bug with the crash message: `TVMError: unknown intrinsic Op(tir.acos)`

@Lunderberg Indeed, the indices are out of the boundary. However, `Segmentation fault` is a dangerous behavior and is often considered a vulnerability. Do we need an isolation mechanism to check...

@Lunderberg Thanks! The `tvm.tir.analysis.OOBChecker()` is very useful. I like it! It successfully identifies the OOB issue and avoids the segfault. But It was disabled by default. Why not enable it...

@xhmelon Thanks for your investigation. Indeed, the Realx IR is invalid and the crash message also gives the correct warning. However, the above Relax IR passes the `verify_well_formed` validation and...

@Lunderberg Thanks for your investigation. Such information help deeply undestand the usage of different transforms. Due to the incomplete documentation of TVM, understand the usage of each transform based on...

@Lunderberg Fixing for fixing the wrong implementation about `assert_structural_equal`. I have another question. For the given IRs in my script, an odd IRs was obtained after using the `RemoveUnusedOutputs` optimization....

For the dead code (i.e., `def scatter_elements()...`), if we keep it and the call `mod = tvm.tir.transform.DefaultGPUSchedule()(mod)`, to execute the relax IR, the test case will crash unexpectedly. However, if...

@Lunderberg Thanks for your fixing. Segfault is a dangerous behavior, fixing it with an error message is a good strategy. Also thanks for your efforts in improving the well-formed checking...

A similar bug occurs as shown below. Based on what I saw. The well-formed checker commonly corrects the return type and shape. However, when the type of relax function return...

After my investigation, I found if we use the `relax.transform.FuseTIR()` before the `relax.build`, the compilation will run well. However, I still haven't figured out why the compilation crash was unexpected...