Merge branch 'numa-improvements-take2' of https://github.com/dbsanfte/llama.cpp into numa-improvements-take2

Author: dbsanfte

ggml/include/ggml.h

@@ -638,7 +638,9 @@ extern "C" {
         void * extra; // extra things e.g. for ggml-cuda.cu
 
 #ifdef GGML_NUMA_MIRROR
-        char padding[4];
+        char padding[10];
+#else
+	char padding[8];
 #endif
     };
 

ggml/src/ggml.c

@@ -1672,9 +1672,7 @@ static struct ggml_tensor * ggml_new_tensor_impl(
 #endif
         /*.name         =*/ { 0 },
         /*.extra        =*/ NULL,
-#ifdef GGML_NUMA_MIRROR
         /*.padding      =*/ { 0 },
-#endif
     };
 
     // TODO: this should not be needed as long as we don't rely on aligned SIMD loads