feat(ck-tile): multi-ABD GEMM TE to dispatcher bridge (#9305)

ISSUE ID: #8997

## Motivation

The CK Tile dispatcher could already generate and launch regular GEMM
through the
TileEngine → Dispatcher bridge, but it had no path for the multi-tensor
**gemm_multi_abd** op. Multi-ABD is used when a GEMM needs to combine
several A and
B operands and fuse several D operands in the epilogue
(`E = cde_op(a_op(As) @ b_op(Bs), {Ds})`), which is a real Old-TE
capability with no
dispatcher equivalent. This PR closes that gap so Python callers can
drive multi_abd
through the dispatcher at parity with the legacy Tile Engine version,
without touching
C++.

It follows the divergent-ABI pattern established by the grouped bridge
(#9000) because
multi_abd needs **arrays** of A/B/D device pointers, not the
single-pointer regular
GEMM ABI. The capability set matches the Old-TE
`gemm_multi_abd_instance_builder.py`
exactly: `fp16`, `rcrr` layout, configurable A/B/D tensor counts, and
the element-wise
op set `{PassThrough, AddScale, MultiDMultiply, MultiDAdd}`.



## Test Plan

- Run the CPU-only unit tests (no GPU required):
  `python3 -m pytest dispatcher/tests/test_multi_abd_bridge.py -v`
- On-GPU numeric verification through the bridge launch path (gfx942 /
MI300X),
512x512x512 fp16 rcrr, across the default 2/2/2 all-PassThrough config
and
  non-PassThrough element-wise ops.
- Confirm the CI and default config expansions yield the expected kernel
counts.

## Test Result

- CPU-only unit tests pass (10 passed).
- Numeric verification (bridge launch path), 512x512x512 fp16 rcrr:
  - default 2/2/2 all-PassThrough: `max_rel = 2.9e-4`
  - CDE = MultiDAdd: `max_rel = 5.7e-4`
  - A-op = MultiDAdd: `max_rel = 4.1e-4`
  - all far below the fp16 tolerance (2e-2); 0 failed measurements.
- CI config expands to 16 arch-valid kernels; `default_config.json` →
8896.
- `standard` variant `expand_sweep` regression clean.
- clang-format-18 (18.1.8) clean on `gemm_multi_abd_ctypes_lib.cpp`.
- Serialized A/B perf-parity vs Old-TE (MI300X / gfx942, fp16 rcrr, 16
stems × 5
shapes = 80 rows, interleaved, fair 50/100/flush/rotating both sides):
**at
parity** — median gap -0.24%, mean -0.66%, 100% within ±15%, 87.5%
within ±5%
  (range [-9.57%, +5.35%]). See the parity comment for details.



---

**Related PRs / references (TileEngine → Dispatcher GEMM bridge
series):** #8997 (regular GEMM fp16/bf16 all-layout), #9000 (grouped
GEMM), #9028 (stream-K), #8887 (fp8/bf8/int8). This PR is a sibling in
the same bridge effort tracked across those PRs.

---------

Co-authored-by: Muhammed Emin Ozturk <3836908+ozturkosu@users.noreply.github.com>

Author: ozturkosu

projects/composablekernel/dispatcher/bindings/ctypes/CMakeLists.txt

@@ -276,6 +276,91 @@ else()
     )
 endif()
 
+# =============================================================================
+# TileEngine -> Dispatcher Bridge ctypes Libraries (append-only shared block)
+# =============================================================================
+#
+# The bridge PRs (#9305 gemm_multi_abd, #9306 batched_gemm, #9328
+# batched_contraction) each add a registry-bypass ctypes .so target. To keep the
+# branches mutually conflict-proof, this block is BYTE-IDENTICAL on every branch
+# that touches this file. Each target is doubly guarded:
+#   (1) on its generated kernel-header glob, AND
+#   (2) on if(EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/<op>_ctypes_lib.cpp)
+# so a branch that does not ship a given <op>_ctypes_lib.cpp simply skips that
+# target at configure time (no error). The Python bridges build these .so files
+# at runtime via hipcc; these CMake targets exist for CMake-driven / CI builds.
+
+# --- GEMM Multi-ABD (registry-bypass, array-pointer ABI) ---------------------
+if(EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/gemm_multi_abd_ctypes_lib.cpp)
+    file(GLOB GEMM_MULTI_ABD_KERNEL_HEADERS
+        "${CMAKE_BINARY_DIR}/generated_kernels/gemm_*_multiabd_*.hpp")
+    if(GEMM_MULTI_ABD_KERNEL_HEADERS)
+        list(SORT GEMM_MULTI_ABD_KERNEL_HEADERS)
+        list(GET GEMM_MULTI_ABD_KERNEL_HEADERS 0 GEMM_MULTI_ABD_KERNEL_HEADER)
+        message(STATUS "Found GEMM Multi-ABD kernel for ctypes lib: ${GEMM_MULTI_ABD_KERNEL_HEADER}")
+        add_ctypes_library(dispatcher_gemm_multi_abd_lib
+            gemm_multi_abd_ctypes_lib.cpp
+            KERNEL_HEADER ${GEMM_MULTI_ABD_KERNEL_HEADER}
+        )
+        target_compile_definitions(dispatcher_gemm_multi_abd_lib PRIVATE
+            CK_TILE_SINGLE_KERNEL_INCLUDE)
+    else()
+        message(STATUS
+            "No GEMM Multi-ABD kernel found for ctypes lib - skipping dispatcher_gemm_multi_abd_lib "
+            "(built at runtime by the Python bridge once a kernel header exists)")
+    endif()
+endif()
+
+# --- Batched GEMM (registry-bypass, batch_count + per-batch strides ABI) ------
+if(EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/batched_gemm_ctypes_lib.cpp)
+    file(GLOB BATCHED_GEMM_KERNEL_HEADERS "${CMAKE_BINARY_DIR}/generated_kernels/gemm_*_batched.hpp")
+    if(BATCHED_GEMM_KERNEL_HEADERS)
+        list(SORT BATCHED_GEMM_KERNEL_HEADERS)
+        list(GET BATCHED_GEMM_KERNEL_HEADERS 0 BATCHED_GEMM_KERNEL_HEADER)
+        message(STATUS "Found Batched GEMM kernel for ctypes lib: ${BATCHED_GEMM_KERNEL_HEADER}")
+        add_library(dispatcher_batched_gemm_lib SHARED batched_gemm_ctypes_lib.cpp)
+        target_include_directories(dispatcher_batched_gemm_lib PRIVATE
+            ${PROJECT_SOURCE_DIR}/include
+            ${PROJECT_SOURCE_DIR}/dispatcher/include
+        )
+        target_link_libraries(dispatcher_batched_gemm_lib PRIVATE hip::device)
+        target_compile_options(dispatcher_batched_gemm_lib PRIVATE
+            -include ${BATCHED_GEMM_KERNEL_HEADER}
+        )
+        target_compile_definitions(dispatcher_batched_gemm_lib PRIVATE CK_TILE_SINGLE_KERNEL_INCLUDE)
+        set_target_properties(dispatcher_batched_gemm_lib PROPERTIES
+            POSITION_INDEPENDENT_CODE ON
+            CXX_STANDARD 17
+        )
+    else()
+        message(STATUS "No Batched GEMM kernel found for ctypes lib - skipping dispatcher_batched_gemm_lib")
+    endif()
+endif()
+
+# --- Batched-Contraction (registry-bypass, force-included kernel) ------------
+if(EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/batched_contraction_ctypes_lib.cpp)
+    file(GLOB BC_KERNEL_HEADERS "${CMAKE_BINARY_DIR}/generated_kernels/batched_contraction_*.hpp")
+    if(BC_KERNEL_HEADERS)
+        list(SORT BC_KERNEL_HEADERS)
+        list(GET BC_KERNEL_HEADERS 0 BC_KERNEL_HEADER)
+        message(STATUS "Found batched-contraction kernel for ctypes lib: ${BC_KERNEL_HEADER}")
+        add_ctypes_library(dispatcher_batched_contraction_lib
+            batched_contraction_ctypes_lib.cpp
+            KERNEL_HEADER ${BC_KERNEL_HEADER}
+        )
+        # The generated header only exports SelectedKernel/KERNEL_NAME/CONTRACTION_KEY_*
+        # under this define, so the force-include path requires it.
+        target_compile_definitions(dispatcher_batched_contraction_lib PRIVATE
+            CK_TILE_SINGLE_KERNEL_INCLUDE
+        )
+    else()
+        message(STATUS
+            "No batched-contraction kernel found for ctypes lib - skipping "
+            "dispatcher_batched_contraction_lib (built at runtime by the Python "
+            "bridge once a kernel header exists)")
+    endif()
+endif()
+
 # =============================================================================
 # GPU Helper Executable
 # =============================================================================