[MetaSchedule][Hexagon] Add postproc for verifying VTCM usage #13538
Conversation
|
Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment.
Generated by tvm-bot |
masahi
changed the title
|
|
||
| def test_conv2d_vtcm(): | ||
| def get_target(vtcm_cap): | ||
| target = tvm.target.hexagon("v68", vtcm_capacity=vtcm_cap) |
There was a problem hiding this comment.
You can use get_hexagon_target (from .infrastructure import get_hexagon_target) instead of writing your own version
There was a problem hiding this comment.
I don't want to depend on a file from contrib/test_hexagon from this file.
| @@ -87,6 +87,15 @@ TVM_REGISTER_GLOBAL("tir.analysis.calculate_allocated_bytes").set_body_typed([]( | |||
| return CalculateAllocatedBytes(func); | |||
| }); | |||
|
|
|||
| bool VerifyVTCMLimit(const PrimFunc& func, Integer limit) { | |||
There was a problem hiding this comment.
looks like transform::VerifyVTCMLimit also uses this logic. Can we reuse this function inside that one?
There was a problem hiding this comment.
Yeah I wanted to do that, but the pass variant needs vtcm_allocated in its error message, while this one wants to return only true / false. Refactoring is possible but I thought just duplicating might be cleaner in this case.
|
LGTM |
…#13538) * add new postproc VerifyVTCMLimit * remove pass * add test * add doc, missing file * Add back VectorizeLoop in prereq lowering pass * fix lint
Building on #13349, and similar to
VerifyGPUCode, add a postproc check for the VTCM usage.@csullivan @Icemist @nverke