Missed optimization on fixed-size bytes comparison

[ordian]

https://play.rust-lang.org/?version=stable&mode=release&edition=2018&gist=99eabb4a672da50556a9eb7d4f4016e9

cc #62531

The generated assembly calles into memcmp, but still is slower.

playground::H256::eq_libc:
	movdqu	(%rdi), %xmm0
	movdqu	16(%rdi), %xmm1
	movdqu	(%rsi), %xmm2
	pcmpeqb	%xmm0, %xmm2
	movdqu	16(%rsi), %xmm0
	pcmpeqb	%xmm1, %xmm0
	pand	%xmm2, %xmm0
	pmovmskb	%xmm0, %eax
	cmpl	$65535, %eax
	sete	%al
	retq

playground::H256::eq_derived:
	cmpq	%rsi, %rdi
	je	.LBB1_1
	movdqu	(%rdi), %xmm0
	movdqu	16(%rdi), %xmm1
	movdqu	(%rsi), %xmm2
	pcmpeqb	%xmm0, %xmm2
	movdqu	16(%rsi), %xmm0
	pcmpeqb	%xmm1, %xmm0
	pand	%xmm2, %xmm0
	pmovmskb	%xmm0, %eax
	cmpl	$65535, %eax
	sete	%al
	retq

.LBB1_1:
	movb	$1, %al
	retq

playground::H256::cmp_libc:
	pushq	%rax
	movl	$32, %edx
	callq	*memcmp@GOTPCREL(%rip)
	xorl	%ecx, %ecx
	testl	%eax, %eax
	setne	%cl
	movl	$255, %eax
	cmovnsl	%ecx, %eax
	popq	%rcx
	retq

playground::H256::cmp_derived:
	pushq	%rax
	movl	$32, %edx
	callq	*memcmp@GOTPCREL(%rip)
	testl	%eax, %eax
	setns	%cl
	addb	%cl, %cl
	addb	$-1, %cl
	testl	%eax, %eax
	movzbl	%cl, %ecx
	cmovnel	%ecx, %eax
	popq	%rcx
	retq