ROCm / ROCm/AMDMIGraphX

[BUG] Buffer shape not the same as Output Shape

Open
#3,334 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

bug
Dominant language
C++
Stars
333
Forks
150
Avg merge
4d 19h
Merged PRs (30d)
54

Description

When offload copy is False, output shape transformations are ignored (just returns content of last buffer #output_0).
Buffer shape is not the same as output shape!

@7187 = gpu::code_object[code_object=4424,symbol_name=convert_kernel,global=40448,local=1024,](@7185,main:#output_0) -> float_type, {79, 1, 1, 1024}, {1024, 1024, 1024, 1}
@7188 = squeeze[axes={1}](@7187) -> float_type, {79, 1, 1024}, {1024, 1024, 1}
@7189 = transpose[permutation={1, 0, 2}](@7188) -> float_type, {1, 79, 1024}, {1024, 1024, 1}
@7190 = @return(@7189,@7104)

{79, 1, 1, 1024} is returned instead of expected {1, 79, 1024}

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by tracing the offload-copy-false path for the graph shown in the issue, focusing on how the final squeeze and transpose output shape is handled versus the last buffer. Reproduce the case and verify that the returned shape is {1, 79, 1024} rather than {79, 1, 1, 1024}.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
backend, machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.