diff --git a/Makefile b/Makefile
index 87fe795aa..804c1bbec 100644
--- a/Makefile
+++ b/Makefile
@@ -359,6 +359,10 @@ ifdef LLAMA_SERVER_SSL
 	MK_LDFLAGS += -lssl -lcrypto
 endif
 
+ifndef GGML_NO_CPU_AARCH64
+	MK_CPPFLAGS += -DGGML_USE_CPU_AARCH64
+endif
+
 # warnings
 WARN_FLAGS = \
 	-Wall \
@@ -940,10 +944,6 @@ ggml/src/ggml-cuda/%.o: \
 	$(MCC) $(CXXFLAGS) $(MUSAFLAGS) -x musa -mtgpu -c -o $@ $<
 endif # GGML_MUSA
 
-ifndef GGML_NO_CPU_AARCH64
-	MK_CPPFLAGS += -DGGML_USE_CPU_AARCH64
-endif
-
 ifdef GGML_METAL
 	MK_CPPFLAGS += -DGGML_USE_METAL
 	MK_LDFLAGS  += -framework Foundation -framework Metal -framework MetalKit
diff --git a/ggml/src/ggml-cpu/CMakeLists.txt b/ggml/src/ggml-cpu/CMakeLists.txt
index 30de6c99a..cef41a074 100644
--- a/ggml/src/ggml-cpu/CMakeLists.txt
+++ b/ggml/src/ggml-cpu/CMakeLists.txt
@@ -143,14 +143,23 @@ elseif (CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64" OR CMAKE_GENERATOR_PLATFORM_LW
             if (GGML_AVX512_VBMI)
                 add_compile_definitions($<$<COMPILE_LANGUAGE:C>:__AVX512VBMI__>)
                 add_compile_definitions($<$<COMPILE_LANGUAGE:CXX>:__AVX512VBMI__>)
+                if (CMAKE_C_COMPILER_ID STREQUAL "Clang")
+                    list(APPEND ARCH_FLAGS -mavx512vbmi)
+                endif()
             endif()
             if (GGML_AVX512_VNNI)
                 add_compile_definitions($<$<COMPILE_LANGUAGE:C>:__AVX512VNNI__>)
                 add_compile_definitions($<$<COMPILE_LANGUAGE:CXX>:__AVX512VNNI__>)
+                if (CMAKE_C_COMPILER_ID STREQUAL "Clang")
+                    list(APPEND ARCH_FLAGS -mavx512vnni)
+                endif()
             endif()
             if (GGML_AVX512_BF16)
                 add_compile_definitions($<$<COMPILE_LANGUAGE:C>:__AVX512BF16__>)
                 add_compile_definitions($<$<COMPILE_LANGUAGE:CXX>:__AVX512BF16__>)
+                if (CMAKE_C_COMPILER_ID STREQUAL "Clang")
+                    list(APPEND ARCH_FLAGS -mavx512bf16)
+                endif()
             endif()
             if (GGML_AMX_TILE)
                 add_compile_definitions($<$<COMPILE_LANGUAGE:C>:__AMX_TILE__>)
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c
index 4a97bfc32..5cdf59f25 100644
--- a/ggml/src/ggml.c
+++ b/ggml/src/ggml.c
@@ -49,6 +49,14 @@
 
 #define UNUSED GGML_UNUSED
 
+#if defined(_MSC_VER)
+#define m512bh(p) p
+#define m512i(p) p
+#else
+#define m512bh(p) (__m512bh)(p)
+#define m512i(p) (__m512i)(p)
+#endif
+
 // precomputed f32 table for f16 (256 KB) (ggml-impl.h)
 float ggml_table_f32_f16[1 << 16];
 
diff --git a/scripts/compare-llama-bench.py b/scripts/compare-llama-bench.py
index 4ac6b5fc0..6125a080a 100755
--- a/scripts/compare-llama-bench.py
+++ b/scripts/compare-llama-bench.py
@@ -19,22 +19,22 @@ logger = logging.getLogger("compare-llama-bench")
 
 # Properties by which to differentiate results per commit:
 KEY_PROPERTIES = [
-    "cpu_info", "gpu_info", "n_gpu_layers", "cuda", "vulkan", "kompute", "metal", "sycl", "rpc", "gpu_blas",
-    "blas", "model_filename", "model_type", "n_batch", "n_ubatch", "embeddings", "n_threads",
-    "type_k", "type_v", "use_mmap", "no_kv_offload", "split_mode", "main_gpu", "tensor_split", "flash_attn", "n_prompt", "n_gen"
+    "cpu_info", "gpu_info", "backends", "n_gpu_layers", "model_filename", "model_type", "n_batch", "n_ubatch",
+    "embeddings", "cpu_mask", "cpu_strict", "poll", "n_threads", "type_k", "type_v", "use_mmap", "no_kv_offload",
+    "split_mode", "main_gpu", "tensor_split", "flash_attn", "n_prompt", "n_gen"
 ]
 
 # Properties that are boolean and are converted to Yes/No for the table:
-BOOL_PROPERTIES = ["cuda", "vulkan", "kompute", "metal", "sycl", "gpu_blas", "blas", "embeddings", "use_mmap", "no_kv_offload", "flash_attn"]
+BOOL_PROPERTIES = ["embeddings", "cpu_strict", "use_mmap", "no_kv_offload", "flash_attn"]
 
 # Header names for the table:
 PRETTY_NAMES = {
-    "cuda": "CUDA", "vulkan": "Vulkan", "kompute": "Kompute", "metal": "Metal", "sycl": "SYCL", "rpc": "RPC",
-    "gpu_blas": "GPU BLAS", "blas": "BLAS", "cpu_info": "CPU", "gpu_info": "GPU", "model_filename": "File", "model_type": "Model",
-    "model_size": "Model Size [GiB]", "model_n_params": "Num. of Par.", "n_batch": "Batch size", "n_ubatch": "Microbatch size",
-    "n_threads": "Threads", "type_k": "K type", "type_v": "V type", "n_gpu_layers": "GPU layers", "split_mode": "Split mode",
-    "main_gpu": "Main GPU", "no_kv_offload": "NKVO", "flash_attn": "FlashAttention", "tensor_split": "Tensor split",
-    "use_mmap": "Use mmap", "embeddings": "Embeddings",
+    "cpu_info": "CPU", "gpu_info": "GPU", "backends": "Backends", "n_gpu_layers": "GPU layers",
+    "model_filename": "File", "model_type": "Model", "model_size": "Model size [GiB]",
+    "model_n_params": "Num. of par.", "n_batch": "Batch size", "n_ubatch": "Microbatch size",
+    "embeddings": "Embeddings", "cpu_mask": "CPU mask", "cpu_strict": "CPU strict", "poll": "Poll",
+    "n_threads": "Threads", "type_k": "K type", "type_v": "V type", "split_mode": "Split mode", "main_gpu": "Main GPU",
+    "no_kv_offload": "NKVO", "flash_attn": "FlashAttention", "tensor_split": "Tensor split", "use_mmap": "Use mmap",
 }
 
 DEFAULT_SHOW = ["model_type"]  # Always show these properties by default.