sub \$8,$len()

in ring/crypto/fipsmodule/aes/asm/aesni-x86_64.pl [968:1154]

• 169 lines of code
• 3 McCabe index (conditional complexity)

	sub	\$8,$len
	jnc	.Lctr32_loop8			# loop if $len-=8 didn't borrow

	add	\$8,$len			# restore real remaining $len
	jz	.Lctr32_done			# done if ($len==0)
	lea	-0x80($key),$key

.Lctr32_tail:
	# note that at this point $inout0..5 are populated with
	# counter values xor-ed with 0-round key
	lea	16($key),$key
	cmp	\$4,$len
	jb	.Lctr32_loop3
	je	.Lctr32_loop4

	# if ($len>4) compute 7 E(counter)
	shl		\$4,$rounds
	movdqa		0x60(%rsp),$inout6
	pxor		$inout7,$inout7

	$movkey		16($key),$rndkey0
	aesenc		$rndkey1,$inout0
	aesenc		$rndkey1,$inout1
	lea		32-16($key,$rounds),$key# prepare for .Lenc_loop8_enter
	neg		%rax
	aesenc		$rndkey1,$inout2
	add		\$16,%rax		# prepare for .Lenc_loop8_enter
	 movups		($inp),$in0
	aesenc		$rndkey1,$inout3
	aesenc		$rndkey1,$inout4
	 movups		0x10($inp),$in1		# pre-load input
	 movups		0x20($inp),$in2
	aesenc		$rndkey1,$inout5
	aesenc		$rndkey1,$inout6

	call            .Lenc_loop8_enter

	movdqu	0x30($inp),$in3
	pxor	$in0,$inout0
	movdqu	0x40($inp),$in0
	pxor	$in1,$inout1
	movdqu	$inout0,($out)			# store output
	pxor	$in2,$inout2
	movdqu	$inout1,0x10($out)
	pxor	$in3,$inout3
	movdqu	$inout2,0x20($out)
	pxor	$in0,$inout4
	movdqu	$inout3,0x30($out)
	movdqu	$inout4,0x40($out)
	cmp	\$6,$len
	jb	.Lctr32_done			# $len was 5, stop store

	movups	0x50($inp),$in1
	xorps	$in1,$inout5
	movups	$inout5,0x50($out)
	je	.Lctr32_done			# $len was 6, stop store

	movups	0x60($inp),$in2
	xorps	$in2,$inout6
	movups	$inout6,0x60($out)
	jmp	.Lctr32_done			# $len was 7, stop store

.align	32
.Lctr32_loop4:
	aesenc		$rndkey1,$inout0
	lea		16($key),$key
	dec		$rounds
	aesenc		$rndkey1,$inout1
	aesenc		$rndkey1,$inout2
	aesenc		$rndkey1,$inout3
	$movkey		($key),$rndkey1
	jnz		.Lctr32_loop4
	aesenclast	$rndkey1,$inout0
	aesenclast	$rndkey1,$inout1
	 movups		($inp),$in0		# load input
	 movups		0x10($inp),$in1
	aesenclast	$rndkey1,$inout2
	aesenclast	$rndkey1,$inout3
	 movups		0x20($inp),$in2
	 movups		0x30($inp),$in3

	xorps	$in0,$inout0
	movups	$inout0,($out)			# store output
	xorps	$in1,$inout1
	movups	$inout1,0x10($out)
	pxor	$in2,$inout2
	movdqu	$inout2,0x20($out)
	pxor	$in3,$inout3
	movdqu	$inout3,0x30($out)
	jmp	.Lctr32_done			# $len was 4, stop store

.align	32
.Lctr32_loop3:
	aesenc		$rndkey1,$inout0
	lea		16($key),$key
	dec		$rounds
	aesenc		$rndkey1,$inout1
	aesenc		$rndkey1,$inout2
	$movkey		($key),$rndkey1
	jnz		.Lctr32_loop3
	aesenclast	$rndkey1,$inout0
	aesenclast	$rndkey1,$inout1
	aesenclast	$rndkey1,$inout2

	movups	($inp),$in0			# load input
	xorps	$in0,$inout0
	movups	$inout0,($out)			# store output
	cmp	\$2,$len
	jb	.Lctr32_done			# $len was 1, stop store

	movups	0x10($inp),$in1
	xorps	$in1,$inout1
	movups	$inout1,0x10($out)
	je	.Lctr32_done			# $len was 2, stop store

	movups	0x20($inp),$in2
	xorps	$in2,$inout2
	movups	$inout2,0x20($out)		# $len was 3, stop store

.Lctr32_done:
	xorps	%xmm0,%xmm0			# clear register bank
	xor	$key0,$key0
	pxor	%xmm1,%xmm1
	pxor	%xmm2,%xmm2
	pxor	%xmm3,%xmm3
	pxor	%xmm4,%xmm4
	pxor	%xmm5,%xmm5
___
$code.=<<___ if (!$win64);
	pxor	%xmm6,%xmm6
	pxor	%xmm7,%xmm7
	movaps	%xmm0,0x00(%rsp)		# clear stack
	pxor	%xmm8,%xmm8
	movaps	%xmm0,0x10(%rsp)
	pxor	%xmm9,%xmm9
	movaps	%xmm0,0x20(%rsp)
	pxor	%xmm10,%xmm10
	movaps	%xmm0,0x30(%rsp)
	pxor	%xmm11,%xmm11
	movaps	%xmm0,0x40(%rsp)
	pxor	%xmm12,%xmm12
	movaps	%xmm0,0x50(%rsp)
	pxor	%xmm13,%xmm13
	movaps	%xmm0,0x60(%rsp)
	pxor	%xmm14,%xmm14
	movaps	%xmm0,0x70(%rsp)
	pxor	%xmm15,%xmm15
___
$code.=<<___ if ($win64);
	movaps	-0xa8($key_),%xmm6
	movaps	%xmm0,-0xa8($key_)		# clear stack
	movaps	-0x98($key_),%xmm7
	movaps	%xmm0,-0x98($key_)
	movaps	-0x88($key_),%xmm8
	movaps	%xmm0,-0x88($key_)
	movaps	-0x78($key_),%xmm9
	movaps	%xmm0,-0x78($key_)
	movaps	-0x68($key_),%xmm10
	movaps	%xmm0,-0x68($key_)
	movaps	-0x58($key_),%xmm11
	movaps	%xmm0,-0x58($key_)
	movaps	-0x48($key_),%xmm12
	movaps	%xmm0,-0x48($key_)
	movaps	-0x38($key_),%xmm13
	movaps	%xmm0,-0x38($key_)
	movaps	-0x28($key_),%xmm14
	movaps	%xmm0,-0x28($key_)
	movaps	-0x18($key_),%xmm15
	movaps	%xmm0,-0x18($key_)
	movaps	%xmm0,0x00(%rsp)
	movaps	%xmm0,0x10(%rsp)
	movaps	%xmm0,0x20(%rsp)
	movaps	%xmm0,0x30(%rsp)
	movaps	%xmm0,0x40(%rsp)
	movaps	%xmm0,0x50(%rsp)
	movaps	%xmm0,0x60(%rsp)
	movaps	%xmm0,0x70(%rsp)
___
$code.=<<___;
	mov	-8($key_),%rbp
.cfi_restore	%rbp
	lea	($key_),%rsp
.cfi_def_cfa_register	%rsp
.Lctr32_epilogue:
	ret
.cfi_endproc
.size	GFp_${PREFIX}_ctr32_encrypt_blocks,.-GFp_${PREFIX}_ctr32_encrypt_blocks