memcpy_erms.patch

diff --git a/kernel/lib/libk/arch/x86_64/memcpy_impl.S b/kernel/lib/libk/arch/x86_64/memcpy_impl.S
index a2366c51..d5bb54ce 100644
--- a/kernel/lib/libk/arch/x86_64/memcpy_impl.S
+++ b/kernel/lib/libk/arch/x86_64/memcpy_impl.S
@@ -129,6 +129,23 @@ L(1_byte):
 
     ALIGN_TEXT
 L(erms):
+    /* Align the dest to 32 bytes (optimal alignment) */
+    /* Hopefully source gets aligned to 32 in the process */
+    movq  (%rsi), %rcx
+    movq 8(%rsi), %r8
+    movq 16(%rsi), %r9
+    movq 24(%rsi), %r10
+    movq %rcx, (%rdi)
+    movq %r8,  8(%rdi)
+    movq %r9,  16(%rdi)
+    movq %r10, 24(%rdi)
+    lea 32(%rdi), %rcx
+    and $-32, %rcx
+    mov %rcx, %r8
+    sub %rdi, %r8
+    mov %rcx, %rdi
+    add %r8, %rsi
+    sub %r8, %rdx
     mov %rdx, %rcx
     rep movsb