cktile weight preshuffle test and auto tuning for a8w8

[solinzby1]

Motivation

Technical Details

Test Plan

Test Result

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[Copilot]

Pull Request Overview

This PR introduces CKTILE-based weight preshuffle functionality for FP8 (a8w8) GEMM operations, including comprehensive test coverage and auto-tuning infrastructure. The implementation provides an alternative kernel backend for quantized matrix multiplication operations on AMD ROCm GPUs.

Key changes:

• New CKTILE-based FP8 GEMM kernel implementation with weight pre-shuffling support
• Auto-tuning infrastructure with 139+ kernel configurations for gfx942 and 133+ for gfx950 architectures
• Python API integration and JIT compilation support for dynamic kernel selection

Reviewed Changes

Copilot reviewed 18 out of 18 changed files in this pull request and generated 21 comments.

Show a summary per file

File	Description
op_tests/test_gemm_a8w8.py	Adds test function for new CKTILE bpreshuffle kernel
csrc/rocm_ops.cpp	Includes new CKTILE header and fixes include ordering
csrc/pybind/*.cu	PyBind11 bindings for main and tuning interfaces
csrc/include/rocm_ops.hpp	Macro definitions for Python binding registration
csrc/cktile_gemm_a8w8_bpreshuffle/*.cu	Core kernel implementation and tuning dispatch logic
csrc/cktile_gemm_a8w8_bpreshuffle/*.cuh	Common utilities and template configurations
csrc/cktile_gemm_a8w8_bpreshuffle/*.py	Code generation and kernel tuning scripts
aiter/ops/gemm_op_a8w8.py	High-level Python API and dispatch functions
aiter/jit/*.py & *.json	JIT compilation configuration
aiter/configs/*.csv	Tuned and untuned kernel configuration databases
csrc/cktile_gemm_a8w8_bpreshuffle/README.md	Usage documentation for tuning workflow

---

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

[valarLip]

let's havee some data compare for deepseek's shape