Using GNU assembler v. GNU assembler version 2.17.50 (i486-linux-gnu)

Intel machine that's P4 or more.

.section .data
values:
.int 105, 235, 61, 315, 134, 221, 53, 145, 117, 5

.section .text
.globl _start
_start:
nop
movl $values, %esi
movl $9, %ecx
movl $9, %ebx
loop:
cmp (%esi), 4(%esi)
jge skip
xchg %eax, 4(%esi)
movl %eax, (%esi)
skip:
add $4, %esi
dec %ebx
jnz loop
dec %ecx
jz end
movl $values, %esi
movl %ecx, %ebx
jmp loop
end:
movl $1, %eax
movl $0, %ebx
int $0x80

Here's what happens when I assemble it.
exchange02.s:29: Error: too many memory references for `cmp'

This points to the line of code in bold.

If I were to replace that one bold line with these two lines, everything works fine.
movl (%esi), %eax
cmp %eax, 4(%esi)

Here's why I think the above is giving me an assemble time error. Essentially the cmp instruction cannot compare two pieces of memory (sayth so the Intel docs.) Now, when the above instruction (in bold) executes, the parentheses around the two operands force the pointers to which the data is pointing to (and reside in the ESI register) de-reference, thereby creating the above problem that the docs say should not happen.

Is my thinking correct? Yes or no?