refactor: move CUTLASS ownership into sd.cpp

.gitmodules

@@ -10,3 +10,6 @@
 [submodule "thirdparty/libwebm"]
 	path = thirdparty/libwebm
 	url = https://github.com/webmproject/libwebm.git
+[submodule "thirdparty/cutlass"]
+	path = thirdparty/cutlass
+	url = https://github.com/NVIDIA/cutlass.git

CMakeLists.txt

@@ -346,12 +346,13 @@ if (SD_CUDA)
     target_link_libraries(${SD_LIB} PUBLIC CUDA::cuda_driver)
     set_property(SOURCE src/core/ggml_extend_backend.cpp APPEND PROPERTY COMPILE_DEFINITIONS SD_USE_CUDA)
     if (NOT SD_USE_UPSTREAM_GGML)
+        include(cmake/cutlass.cmake)
         enable_language(CUDA)
         set_target_properties(${SD_LIB} PROPERTIES CUDA_STANDARD 17 CUDA_STANDARD_REQUIRED ON)
-        if (GGML_CUDA_CUTLASS_PATH)
+        if (SD_CUDA_CUTLASS_PATH)
             target_sources(${SD_LIB} PRIVATE src/core/cuda/cute_attention.cu src/core/cuda/i8_projection.cu src/core/cuda/i8_gated_residual.cu src/core/cuda/i8_gemm.cu)
             set_source_files_properties(src/core/cuda/cute_attention.cu src/core/cuda/i8_projection.cu src/core/cuda/i8_gated_residual.cu src/core/cuda/i8_gemm.cu PROPERTIES
-                INCLUDE_DIRECTORIES "${GGML_CUDA_CUTLASS_PATH}/include"
+                INCLUDE_DIRECTORIES "${SD_CUDA_CUTLASS_PATH}/include"
                 COMPILE_OPTIONS "--use_fast_math;--extended-lambda;--expt-relaxed-constexpr")
             if (MSVC)
                 set_property(SOURCE src/core/cuda/cute_attention.cu src/core/cuda/i8_projection.cu src/core/cuda/i8_gated_residual.cu src/core/cuda/i8_gemm.cu APPEND PROPERTY COMPILE_OPTIONS -Xcompiler=/Zc:preprocessor,/Zc:__cplusplus)

cmake/cutlass.cmake

@@ -0,0 +1,20 @@
+set(sd_cutlass_default "${PROJECT_SOURCE_DIR}/thirdparty/cutlass")
+if(NOT DEFINED SD_CUDA_CUTLASS_PATH AND DEFINED GGML_CUDA_CUTLASS_PATH)
+    get_filename_component(sd_old_cutlass_default "${SD_GGML_SOURCE_DIR}/src/ggml-cuda/vendors/cutlass" ABSOLUTE)
+    if(NOT GGML_CUDA_CUTLASS_PATH STREQUAL sd_old_cutlass_default)
+        set(sd_cutlass_default "${GGML_CUDA_CUTLASS_PATH}")
+    endif()
+    message(STATUS "Migrating GGML_CUDA_CUTLASS_PATH to SD_CUDA_CUTLASS_PATH")
+endif()
+set(SD_CUDA_CUTLASS_PATH "${sd_cutlass_default}" CACHE PATH "sd: CUTLASS source directory for INT8 GEMM and CuTe attention, empty to disable")
+unset(GGML_CUDA_CUTLASS_PATH CACHE)
+if(SD_CUDA_CUTLASS_PATH)
+    if(NOT EXISTS "${SD_CUDA_CUTLASS_PATH}/include/cutlass/gemm/kernel/default_gemm_universal_with_visitor.h" OR
+       NOT EXISTS "${SD_CUDA_CUTLASS_PATH}/include/cute/tensor.hpp")
+        message(FATAL_ERROR
+            "CUTLASS headers not found at ${SD_CUDA_CUTLASS_PATH}. "
+            "Run git submodule update --init --recursive from the sd.cpp repository root, "
+            "set SD_CUDA_CUTLASS_PATH to a CUTLASS source tree (tested with 3.9.2), "
+            "or set SD_CUDA_CUTLASS_PATH to an empty string to disable CUTLASS.")
+    endif()
+endif()

docs/build.md

@@ -111,7 +111,7 @@ cmake .. -DSD_CUDA=ON
 cmake --build . --config Release
 ```
 
-CUDA builds with the bundled patched GGML use the CUTLASS submodule by default
+CUDA extensions use sd.cpp's `thirdparty/cutlass` submodule by default
 for eligible INT8 GEMM and CuTe attention operations. If its headers are missing,
 CMake stops with an initialization hint; it does not download them automatically.
 Run `git submodule update --init --recursive` from the repository root before
@@ -121,14 +121,17 @@ To build with the existing CUDA implementations without CUTLASS, configure from
 the repository root with:
 
 ```shell
-cmake -S . -B build -DSD_CUDA=ON -DGGML_CUDA_CUTLASS_PATH:PATH=
+cmake -S . -B build -DSD_CUDA=ON -DSD_CUDA_CUTLASS_PATH:PATH=
 cmake --build build --config Release
 ```
 
-Alternatively, set `GGML_CUDA_CUTLASS_PATH` to another local CUTLASS source
+Alternatively, set `SD_CUDA_CUTLASS_PATH` to another local CUTLASS source
 directory (tested with 3.9.2). Existing CMake caches retain their configured path;
-pass `-UGGML_CUDA_CUTLASS_PATH` when reconfiguring to restore the bundled default.
+pass `-USD_CUDA_CUTLASS_PATH` when reconfiguring to restore the bundled default.
-Non-CUDA builds do not require CUTLASS.
+Existing `GGML_CUDA_CUTLASS_PATH` caches are migrated on first configuration:
+the old bundled path becomes `thirdparty/cutlass`, while custom paths and an
+empty disabled value are preserved. An explicitly set `SD_CUDA_CUTLASS_PATH`
+takes precedence. Non-CUDA builds do not require CUTLASS.
 
 ## Build with HipBLAS
 

src/core/cuda/CMakeLists.txt

@@ -26,7 +26,7 @@ if(SD_CUDA)
     target_sources(${SD_CUDA_TARGET} PRIVATE sage_attention.cu sol_attention.cu)
     set_source_files_properties(sage_attention.cu sol_attention.cu PROPERTIES COMPILE_OPTIONS "--use_fast_math")
     set_source_files_properties(${SD_CUDA_SOURCES} PROPERTIES COMPILE_OPTIONS "--use_fast_math")
-    if(GGML_CUDA_CUTLASS_PATH)
+    if(SD_CUDA_CUTLASS_PATH)
         set_property(SOURCE "${CMAKE_CURRENT_BINARY_DIR}/int8_cuda.cu" APPEND PROPERTY COMPILE_DEFINITIONS SD_CUDA_I8_CUTLASS)
         set_property(SOURCE "${CMAKE_CURRENT_BINARY_DIR}/registration_cuda.cu" APPEND PROPERTY COMPILE_DEFINITIONS SD_CUDA_CUTE_ATTENTION)
         set_property(SOURCE "${CMAKE_CURRENT_BINARY_DIR}/int8_cuda.cu" APPEND PROPERTY INCLUDE_DIRECTORIES "${PROJECT_SOURCE_DIR}/src/core/cuda")