fix(gemm_a8w8_bpreshuffle): pass splitK/KBatch to CK kernels

[AviralGoelAMD]

Summary

• The splitK parameter was accepted by the tune entry points and KBatch = 2^splitK was computed, but never forwarded to the CK/CKTile kernels — it was hardcoded to 1 in both paths
• Thread KBatch through the full dispatch chain: tune.cu → dispatch → generated kernel wrapper → impl → CK MakeArgument / CKTile args.k_batch
• Inference (non-tune) paths are unaffected via int KBatch = 1 default parameter

[github-actions]

🏷️ CI Guide

Runs automatically on every PR:

• ✅ Pre-checks (submodule verification, code formatting)
• ✅ Aiter op tests (gfx942 + gfx950)
• ✅ Triton tests (only when aiter/ops/triton/** or related paths are changed)

Extended tests (opt-in via labels):

Label	Tests
ci:sglang	SGLang integration tests
ci:atom	ATOM benchmark (DeepSeek-R1 + GPT-OSS)
ci:vllm	vLLM benchmark
ci:all	All of the above

Add labels via the sidebar or gh pr edit 2335 --add-label <label>

[Copilot]

Pull request overview

This PR introduces a KBatch (split-K batch) parameter plumbing into the FP8 a8w8 bpreshuffle GEMM implementations for both CK and CKTile backends, enabling the tuning path to run kernels with split-K.

Changes:

• Add KBatch parameter to CK/CKTile GEMM implementation entrypoints and pass it into kernel args.
• Update code generation (gen_instances.py) so generated kernel wrappers accept KBatch and forward it to the impl.
• Update tune dispatchers to pass the computed KBatch into the selected kernel.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
csrc/cktile_gemm_a8w8_bpreshuffle/include/gemm_a8w8_bpreshuffle_cktile_common.cuh	Add KBatch argument and plumb into args.k_batch.
csrc/cktile_gemm_a8w8_bpreshuffle/gen_instances.py	Generate kernel wrappers/manifest declarations with KBatch.
csrc/cktile_gemm_a8w8_bpreshuffle/gemm_a8w8_bpreshuffle_cktile_tune.cu	Update dispatch function types/calls to pass KBatch.
csrc/ck_gemm_a8w8_bpreshuffle/include/gemm_a8w8_bpreshuffle_common.cuh	Add KBatch argument and plumb into CK MakeArgument.
csrc/ck_gemm_a8w8_bpreshuffle/gen_instances.py	Generate kernel wrappers/manifest declarations with KBatch.
csrc/ck_gemm_a8w8_bpreshuffle/gemm_a8w8_bpreshuffle_tune.cu	Update dispatch function types/calls to pass KBatch.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[AviralGoelAMD]

I;ve resolved all the copilot comments.

[yzhou103]

I think the non-tune path should keep same with tuned path. If a splitK >1 pattern is the best config, the splitK parameter should also be passed to ck kernels