Skip rep movsb in copy_backward if possible

There is currently no measureable performance difference in benchmarks
but it likely will make a difference in real workloads.

library/compiler-builtins/src/mem/x86_64.rs

@@ -73,15 +73,21 @@ pub unsafe fn copy_backward(dest: *mut u8, src: *const u8, count: usize) {
     // We can't separate this block due to std/cld
     asm!(
         "std",
+        "test %ecx, %ecx",
+        "jz 1f",
         "rep movsb",
+        "1:",
         "sub $7, %rsi",
         "sub $7, %rdi",
         "mov {qword_count}, %rcx",
         "rep movsq",
+        "test {pre_byte_count:e}, {pre_byte_count:e}",
+        "jz 1f",
         "add $7, %rsi",
         "add $7, %rdi",
         "mov {pre_byte_count:e}, %ecx",
         "rep movsb",
+        "1:",
         "cld",
         pre_byte_count = in(reg) pre_byte_count,
         qword_count = in(reg) qword_count,