Remove HipW4A16SkinnyLinearKernel and no-group wvSplitK_int4#869
Merged
mgehre-amd merged 3 commits intogfx11from Apr 17, 2026
Merged
Remove HipW4A16SkinnyLinearKernel and no-group wvSplitK_int4#869mgehre-amd merged 3 commits intogfx11from
mgehre-amd merged 3 commits intogfx11from
Conversation
mgehre-amd
changed the title
eble-amd
approved these changes
Apr 13, 2026
eble-amd
left a comment
eble-amd
left a comment
There was a problem hiding this comment.
This is my favorite kind of change. Please merge it as soon as you receive all appropriate approval.
No real INT4 quantized model uses per-channel (group_size=-1) quantization. All AWQ, GPTQ, and compressed-tensors models use per-group scales with group_size=32 or 128, handled by wvSplitK_int4_g. Removing the unused no-groups variant reduces compile time and code complexity. Changes: - Removes ~260 lines of C++ kernel entry point and sweep code - Eliminates group_size=-1 dispatch path in Python kernel wrapper - Keeps shared kernel templates (used by grouped variant via template param) Signed-off-by: Matthias Gehre <matthias.gehre@amd.com>
HybridW4A16LinearKernel fully supersedes it — it handles both the skinny decode path (wvSplitK_int4_g) and large-batch prefill (Triton), including asymmetric/zero-point models. The skinny-only kernel class added no unique capability. Signed-off-by: Matthias Gehre <matthias.gehre@amd.com>
mgehre-amd
force-pushed
the
matthias.remove-wvsplitk-int4-nogrouped
branch
from
055ceeb to
7ba0080
Compare
The test_triton_w4a16_skinny_fmt_gemm_matches_reference test was failing with 1/16384 elements marginally exceeding atol=1e-2 (actual diff: 0.0106). Relax to atol=5e-2, matching the asymmetric test's tolerance on the same kernel. Signed-off-by: Matthias Gehre <matthias.gehre@amd.com>
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
2 participants
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.
Remove wvSplitK_int4 no-groups kernel variant and remove HipW4A16SkinnyLinearKernel. Both are not unused (all int4 model use groups), and HybridW4A16 supersedes HipW4A16SkinnyLinearKernel.