rocm: add hipBLASLt 7.14.0 dat.zlib overlay support - #14
Open
dannybaths wants to merge 1 commit into
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This PR extends
rocm/scripts/hipblaslt_patch_setup.shto support ROCm 7.14.0.ROCm 7.14.0 stores the relevant Tensile catalogue leaves as
*.dat.zlibundergfx942/, so the patcher now detects compressed leaves, decompresses them formsgpack editing, and recompresses them when writing the overlay.
The 7.14.0 case patches only the fp16 forward leaf:
TensileLibrary_HH_HH_HA_Bias_SAV_UA_Type_HH_HPA_Contraction_l_Alik_Bljk_Cijk_Dijk_gfx942.dat.zlibThe 120 small-N fp16 shapes are routed to existing in-pool solution
243113.No kernel binary is built or changed. fp32 and bf16 pass stock on 7.14.0, and
the CPX WorkGroupMappingXCC rewrite is skipped for 7.14.0 because that defect is
already fixed upstream for this release.
Why
On MI300A, stock ROCm 7.14.0 misses the fp16 forward small-N family and returns
zero solutions for the 120-shape sweep. The existing Dock patcher handled older
plain
.datlayouts, but not the 7.14.0.dat.zliblayout.This makes the proven overlay reproducible through the normal Dock packaging
path instead of requiring a manual SDK edit.
Validation
sudo. It produced one patched
.dat.zlibleaf, byte-identical to theempirically proven overlay leaf, md5
3caa7a436470c3aa7a75822504ed4c0d.active.
overlay active.
environment reached the compute node and
Pytorch_HIPBLASLt_Regression_Checkpassed.
19668onPPAC_MI300A_SPXwith the PR rocm: add hipBLASLt 7.14.0 dat.zlib overlay support #14 scratch overlay active.Pytorch_HIPBLASLt_Regression_Checkpassed for fp16, fp32, and bf16; no sudo, live modulefile, SDK, or/optedits were made.