Vulkan Intel Fixes, Optimizations and Debugging Flags (ggerganov#5301)

* Fix Vulkan on Intel ARC

Optimize matmul for Intel ARC

Add Vulkan dequant test

* Add Vulkan debug and validate flags to Make and CMakeLists.txt

* Enable asynchronous transfers in Vulkan backend

* Fix flake8

* Disable Vulkan async backend functions for now

* Also add Vulkan run tests command to Makefile and CMakeLists.txt

CMakeLists.txt

@@ -100,6 +100,10 @@ option(LLAMA_HIPBLAS                         "llama: use hipBLAS"
 option(LLAMA_HIP_UMA                         "llama: use HIP unified memory architecture"       OFF)
 option(LLAMA_CLBLAST                         "llama: use CLBlast"                               OFF)
 option(LLAMA_VULKAN                          "llama: use Vulkan"                                OFF)
+option(LLAMA_VULKAN_CHECK_RESULTS            "llama: run Vulkan op checks"                      OFF)
+option(LLAMA_VULKAN_DEBUG                    "llama: enable Vulkan debug output"                OFF)
+option(LLAMA_VULKAN_VALIDATE                 "llama: enable Vulkan validation"                  OFF)
+option(LLAMA_VULKAN_RUN_TESTS                "llama: run Vulkan tests"                          OFF)
 option(LLAMA_METAL                           "llama: use Metal"                                 ${LLAMA_METAL_DEFAULT})
 option(LLAMA_METAL_NDEBUG                    "llama: disable Metal debugging"                   OFF)
 option(LLAMA_METAL_SHADER_DEBUG              "llama: compile Metal with -fno-fast-math"         OFF)
@@ -431,6 +435,22 @@ if (LLAMA_VULKAN)
 
         add_compile_definitions(GGML_USE_VULKAN)
 
+        if (LLAMA_VULKAN_CHECK_RESULTS)
+            target_compile_definitions(ggml-vulkan PRIVATE GGML_VULKAN_CHECK_RESULTS)
+        endif()
+
+        if (LLAMA_VULKAN_DEBUG)
+            target_compile_definitions(ggml-vulkan PRIVATE GGML_VULKAN_DEBUG)
+        endif()
+
+        if (LLAMA_VULKAN_VALIDATE)
+            target_compile_definitions(ggml-vulkan PRIVATE GGML_VULKAN_VALIDATE)
+        endif()
+
+        if (LLAMA_VULKAN_RUN_TESTS)
+            target_compile_definitions(ggml-vulkan PRIVATE GGML_VULKAN_RUN_TESTS)
+        endif()
+
         set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} ggml-vulkan)
     else()
         message(WARNING "Vulkan not found")

Makefile

@@ -457,6 +457,18 @@ ifdef LLAMA_VULKAN_CHECK_RESULTS
 	MK_CPPFLAGS  += -DGGML_VULKAN_CHECK_RESULTS
 endif
 
+ifdef LLAMA_VULKAN_DEBUG
+	MK_CPPFLAGS  += -DGGML_VULKAN_DEBUG
+endif
+
+ifdef LLAMA_VULKAN_VALIDATE
+	MK_CPPFLAGS  += -DGGML_VULKAN_VALIDATE
+endif
+
+ifdef LLAMA_VULKAN_RUN_TESTS
+	MK_CPPFLAGS  += -DGGML_VULKAN_RUN_TESTS
+endif
+
 ggml-vulkan.o: ggml-vulkan.cpp ggml-vulkan.h
 	$(CXX) $(CXXFLAGS) -c $< -o $@
 endif # LLAMA_VULKAN