[Kernel] (1/N) Machete - Hopper Optimized Mixed Precision Linear Kern…

…el (vllm-project#7174)

.gitignore

@@ -87,6 +87,9 @@ target/
 profile_default/
 ipython_config.py
 
+# generated files
+**/generated/**
+
 # pyenv
 #   For a library or package, you might want to ignore these files since the code is
 #   intended to run in multiple environments; otherwise, check them in:

CMakeLists.txt

@@ -227,6 +227,46 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
           "-gencode arch=compute_90a,code=sm_90a")
   endif()
 
+  #
+  # For the Machete kernels we automatically generate sources for various 
+  # preselected input type pairs and schedules.
+  # Generate sources:
+  execute_process(
+    COMMAND ${CMAKE_COMMAND} -E env 
+    PYTHONPATH=${CMAKE_CURRENT_SOURCE_DIR}/csrc/cutlass_extensions/:${CUTLASS_DIR}/python/:$PYTHONPATH 
+      ${Python_EXECUTABLE} ${CMAKE_CURRENT_SOURCE_DIR}/csrc/quantization/machete/generate.py
+    RESULT_VARIABLE machete_generation_result
+    OUTPUT_VARIABLE machete_generation_output
+    OUTPUT_FILE ${CMAKE_CURRENT_BINARY_DIR}/machete_generation.log
+    ERROR_FILE ${CMAKE_CURRENT_BINARY_DIR}/machete_generation.log
+  )
+
+  if (NOT machete_generation_result EQUAL 0)
+    message(FATAL_ERROR "Machete generation failed."
+                        " Result: \"${machete_generation_result}\"" 
+                        "\nCheck the log for details: "
+                        "${CMAKE_CURRENT_BINARY_DIR}/machete_generation.log")
+  else()
+    message(STATUS "Machete generation completed successfully.")
+  endif()
+
+  # Add machete generated sources
+  file(GLOB MACHETE_GEN_SOURCES "csrc/quantization/machete/generated/*.cu")
+  list(APPEND VLLM_EXT_SRC ${MACHETE_GEN_SOURCES})
+  message(STATUS "Machete generated sources: ${MACHETE_GEN_SOURCES}")
+
+  # See comment above for scaled_mm_c3x (same if condition)
+  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.0)
+    set_source_files_properties(
+          ${MACHETE_GEN_SOURCES}
+          PROPERTIES
+          COMPILE_FLAGS
+          "-gencode arch=compute_90a,code=sm_90a")
+  endif()
+
+  # Add pytorch binding
+  list(APPEND VLLM_EXT_SRC
+    csrc/quantization/machete/machete_pytorch.cu)
 endif()
 
 define_gpu_extension_target(