https://twitter.com/sigfpe/status/584052837732126721

output.asm

	.text
	.file	"fact.c"
	.section	.rodata.cst16,"aM",@progbits,16
	.align	16
.LCPI0_0:
	.long	1                       # 0x1
	.long	1                       # 0x1
	.long	1                       # 0x1
	.long	1                       # 0x1
.LCPI0_1:
	.long	0                       # 0x0
	.long	4294967295              # 0xffffffff
	.long	4294967294              # 0xfffffffe
	.long	4294967293              # 0xfffffffd
.LCPI0_2:
	.long	4294967292              # 0xfffffffc
	.long	4294967291              # 0xfffffffb
	.long	4294967290              # 0xfffffffa
	.long	4294967289              # 0xfffffff9
	.text
	.globl	fact
	.align	16, 0x90
	.type	fact,@function
fact:                                   # @fact
	.cfi_startproc
# BB#0:
	push	rbx
.Ltmp0:
	.cfi_def_cfa_offset 16
.Ltmp1:
	.cfi_offset rbx, -16
                                        # kill: EDI<def> EDI<kill> RDI<def>
	mov	eax, 1
	cmp	edi, 2
	jl	.LBB0_20
# BB#1:                                 # %tailrecurse.preheader
	mov	esi, edi
	not	esi
	cmp	esi, -4
	mov	r10d, -3
	mov	eax, -3
	cmovg	eax, esi
	add	eax, edi
	mov	edx, 1
	cmp	eax, -2
	je	.LBB0_13
# BB#2:                                 # %overflow.checked
	add	eax, 2
	mov	ebx, eax
	and	ebx, -8
	mov	r8d, eax
	and	r8d, -8
	je	.LBB0_3
# BB#4:                                 # %vector.body.preheader
	cmp	esi, -4
	mov	ecx, -3
	cmovg	ecx, esi
	lea	ecx, [rdi + rcx + 2]
	and	ecx, -8
	add	ecx, -8
	shr	ecx, 3
	lea	edx, [rcx + 1]
	test	dl, 1
	je	.LBB0_5
# BB#6:                                 # %vector.body.prol
	movd	xmm0, edi
	pshufd	xmm2, xmm0, 0           # xmm2 = xmm0[0,0,0,0]
	movdqa	xmm3, xmmword ptr [rip + .LCPI0_1] # xmm3 = [0,4294967295,4294967294,4294967293]
	paddd	xmm3, xmm2
	paddd	xmm2, xmmword ptr [rip + .LCPI0_2]
	mov	r11d, 8
	movdqa	xmm1, xmm3
	movdqa	xmm0, xmm2
	jmp	.LBB0_7
.LBB0_3:
	xor	r8d, r8d
	movdqa	xmm3, xmmword ptr [rip + .LCPI0_0] # xmm3 = [1,1,1,1]
	movdqa	xmm2, xmm3
	jmp	.LBB0_12
.LBB0_5:
	movdqa	xmm1, xmmword ptr [rip + .LCPI0_0] # xmm1 = [1,1,1,1]
	xor	r11d, r11d
	pxor	xmm2, xmm2
	pxor	xmm3, xmm3
	movdqa	xmm0, xmm1
.LBB0_7:                                # %vector.body.preheader.split
	mov	r9d, edi
	sub	r9d, ebx
	test	ecx, ecx
	jne	.LBB0_9
# BB#8:
	mov	edi, r9d
	jmp	.LBB0_12
.LBB0_9:                                # %vector.body.preheader.split.split
	mov	edx, edi
	sub	edx, r11d
	cmp	esi, -4
	mov	ecx, -3
	cmovg	ecx, esi
	lea	esi, [rdi + rcx + 2]
	and	esi, -8
	sub	esi, r11d
	movdqa	xmm8, xmmword ptr [rip + .LCPI0_1] # xmm8 = [0,4294967295,4294967294,4294967293]
	movdqa	xmm3, xmmword ptr [rip + .LCPI0_2] # xmm3 = [4294967292,4294967291,4294967290,4294967289]
	.align	16, 0x90
.LBB0_10:                               # %vector.body
                                        # =>This Inner Loop Header: Depth=1
	movd	xmm4, edx
	pshufd	xmm4, xmm4, 0           # xmm4 = xmm4[0,0,0,0]
	movdqa	xmm5, xmm4
	paddd	xmm5, xmm8
	paddd	xmm4, xmm3
	pshufd	xmm6, xmm5, -11         # xmm6 = xmm5[1,1,3,3]
	pmuludq	xmm5, xmm1
	pshufd	xmm7, xmm1, -11         # xmm7 = xmm1[1,1,3,3]
	pmuludq	xmm7, xmm6
	shufps	xmm5, xmm7, -120        # xmm5 = xmm5[0,2],xmm7[0,2]
	shufps	xmm5, xmm5, -40         # xmm5 = xmm5[0,2,1,3]
	pshufd	xmm1, xmm4, -11         # xmm1 = xmm4[1,1,3,3]
	pmuludq	xmm4, xmm0
	pshufd	xmm6, xmm0, -11         # xmm6 = xmm0[1,1,3,3]
	pmuludq	xmm6, xmm1
	shufps	xmm4, xmm6, -120        # xmm4 = xmm4[0,2],xmm6[0,2]
	shufps	xmm4, xmm4, -40         # xmm4 = xmm4[0,2,1,3]
	lea	ecx, [rdx - 8]
	movd	xmm0, ecx
	pshufd	xmm0, xmm0, 0           # xmm0 = xmm0[0,0,0,0]
	movdqa	xmm1, xmm0
	paddd	xmm1, xmm8
	paddd	xmm0, xmm3
	pshufd	xmm2, xmm1, -11         # xmm2 = xmm1[1,1,3,3]
	pmuludq	xmm2, xmm7
	pmuludq	xmm1, xmm5
	shufps	xmm1, xmm2, -120        # xmm1 = xmm1[0,2],xmm2[0,2]
	shufps	xmm1, xmm1, -40         # xmm1 = xmm1[0,2,1,3]
	pshufd	xmm2, xmm0, -11         # xmm2 = xmm0[1,1,3,3]
	pmuludq	xmm2, xmm6
	pmuludq	xmm0, xmm4
	shufps	xmm0, xmm2, -120        # xmm0 = xmm0[0,2],xmm2[0,2]
	shufps	xmm0, xmm0, -40         # xmm0 = xmm0[0,2,1,3]
	add	edx, -16
	add	esi, -16
	jne	.LBB0_10
# BB#11:
	mov	edi, r9d
	movaps	xmm3, xmm1
	movaps	xmm2, xmm0
.LBB0_12:                               # %middle.block
	pshufd	xmm0, xmm2, -11         # xmm0 = xmm2[1,1,3,3]
	pmuludq	xmm2, xmm3
	pshufd	xmm1, xmm3, -11         # xmm1 = xmm3[1,1,3,3]
	pmuludq	xmm1, xmm0
	movdqa	xmm0, xmm2
	shufps	xmm0, xmm1, -120        # xmm0 = xmm0[0,2],xmm1[0,2]
	shufps	xmm0, xmm0, -40         # xmm0 = xmm0[0,2,1,3]
	punpckhdq	xmm2, xmm1      # xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
	pmuludq	xmm2, xmm0
	pshufd	xmm0, xmm1, 78          # xmm0 = xmm1[2,3,0,1]
	pmuludq	xmm0, xmm1
	shufps	xmm2, xmm0, -120        # xmm2 = xmm2[0,2],xmm0[0,2]
	shufps	xmm2, xmm2, -40         # xmm2 = xmm2[0,2,1,3]
	pmuludq	xmm2, xmm0
	pmuludq	xmm0, xmm0
	shufps	xmm2, xmm0, -120        # xmm2 = xmm2[0,2],xmm0[0,2]
	shufps	xmm2, xmm2, -40         # xmm2 = xmm2[0,2,1,3]
	movd	edx, xmm2
	cmp	eax, r8d
	mov	eax, edx
	je	.LBB0_20
.LBB0_13:                               # %tailrecurse.preheader9
	mov	ecx, edi
	not	ecx
	cmp	ecx, -4
	cmovg	r10d, ecx
	lea	esi, [rdi + r10 + 2]
	lea	r8d, [rdi + r10 + 1]
	xor	eax, eax
	test	sil, 7
	je	.LBB0_14
# BB#15:                                # %tailrecurse.prol.preheader
	cmp	ecx, -4
	mov	eax, -3
	cmovg	eax, ecx
	lea	esi, [rdi + rax + 2]
	and	esi, 7
	neg	esi
	mov	eax, edx
	.align	16, 0x90
.LBB0_16:                               # %tailrecurse.prol
                                        # =>This Inner Loop Header: Depth=1
	lea	ecx, [rdi - 1]
	imul	eax, edi
	inc	esi
	mov	edi, ecx
	mov	edx, eax
	jne	.LBB0_16
	jmp	.LBB0_17
.LBB0_14:
	mov	ecx, edi
.LBB0_17:                               # %tailrecurse.preheader9.split
	cmp	r8d, 7
	jb	.LBB0_20
# BB#18:                                # %tailrecurse.preheader9.split.split
	add	ecx, -3
	mov	eax, edx
	.align	16, 0x90
.LBB0_19:                               # %tailrecurse
                                        # =>This Inner Loop Header: Depth=1
	lea	edx, [rcx + 3]
	imul	edx, eax
	lea	eax, [rcx + 2]
	imul	eax, edx
	lea	edx, [rcx + 1]
	imul	edx, eax
	imul	edx, ecx
	lea	eax, [rcx - 1]
	imul	eax, edx
	lea	edx, [rcx - 2]
	imul	edx, eax
	lea	eax, [rcx - 3]
	imul	eax, edx
	lea	edx, [rcx - 4]
	imul	eax, edx
	add	ecx, -8
	cmp	edx, 2
	jg	.LBB0_19
.LBB0_20:                               # %tailrecurse._crit_edge
	pop	rbx
	ret
.Ltmp2:
	.size	fact, .Ltmp2-fact
	.cfi_endproc


	.ident	"clang version 3.6.0 (tags/RELEASE_360/final)"
	.section	".note.GNU-stack","",@progbits