Add cluster launch support to water backend and runtime

[Hardcode84]

• use gpu.launch_func cluster_dims args. during codegen.
• Update GPUToGPURuntime.cpp pass and runtime.
• lit and e2e tests.

[Copilot]

Pull request overview

This PR adds cluster launch support to the water backend and runtime, enabling GPU kernels to use cluster dimensions when launching kernels on AMD GPUs.

Changes:

• Extended the wave_launch_kernel function signature to accept cluster dimension parameters (cluster_x, cluster_y, cluster_z)
• Implemented conditional cluster launch logic using hipDrvLaunchKernelEx when cluster dimensions are specified, falling back to regular hipModuleLaunchKernel otherwise
• Updated the MLIR GPU-to-GPU-Runtime pass to propagate cluster size attributes from gpu.launch_func operations to the runtime

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
wave_lang/kernel/wave/execution_engine/wave_hip_runtime.h	Added cluster dimension parameters (cluster_x, cluster_y, cluster_z) to the wave_launch_kernel function signature
wave_lang/kernel/wave/execution_engine/wave_hip_runtime.cpp	Implemented cluster launch logic using hipDrvLaunchKernelEx with cluster attributes, including optional loading of the function pointer and fallback to regular launch when cluster dimensions are not specified
wave_lang/kernel/compiler/wave_codegen/emitter.py	Added cluster_size generation from hardware constraints and passed it to gpu_d.launch_func, removed unused async_dependencies parameter
water/lib/Transforms/GPUToGPURuntime.cpp	Extended wave_launch_kernel function signature with cluster dimension parameters, added logic to extract and pass cluster dimensions (defaulting to 0) from gpu.launch_func operations
water/test/Transforms/gpu-to-gpu-runtime.mlir	Added test case for cluster launch and updated existing test expectations to account for new cluster dimension parameters
tests/kernel/wave_gemm_test.py	Added use_water_backend parameter to testTensorLoadToShared to enable testing with water backend
lit_tests/kernel/wave/water_host_wrapper.py	Added test_cluster_dims test case to verify cluster dimensions are properly emitted in gpu.launch_func

---

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