Keshav Kini found this example. /* The original code: */ unsigned factorial (unsigned n) { return n == 0 ? 1 : n * factorial(n - 1); } ---------------------------------------------------------------------- # Compiled with gcc, Optimized with -Og .file "fact4.c" .text .globl factorial .type factorial, @function factorial: .LFB0: .cfi_startproc pushq %rbx .cfi_def_cfa_offset 16 .cfi_offset 3, -16 movl %edi, %ebx testl %edi, %edi je .L3 leal -1(%rdi), %edi call factorial imull %ebx, %eax jmp .L2 .L3: movl $1, %eax .L2: popq %rbx .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE0: .size factorial, .-factorial .ident "GCC: (Ubuntu 4.8.4-2ubuntu1~14.04.3) 4.8.4" .section .note.GNU-stack,"",@progbits ---------------------------------------------------------------------- # Compiled with clang, optimized with -O1: .text .file "/tmp/baz.c" .globl factorial .align 16, 0x90 .type factorial,@function factorial: # @factorial .cfi_startproc # BB#0: movl $1, %eax testl %edi, %edi je .LBB0_2 .align 16, 0x90 .LBB0_1: # %tailrecurse # =>This Inner Loop Header: Depth=1 imull %edi, %eax decl %edi jne .LBB0_1 .LBB0_2: # %tailrecurse._crit_edge retq .Ltmp0: .size factorial, .Ltmp0-factorial .cfi_endproc .ident "Debian clang version 3.5.2-2 (tags/RELEASE_352/final) (based on LLVM 3.5.2)" .section ".note.GNU-stack","",@progbits ---------------------------------------------------------------------- # Compiled with clang, optimized with -O2: # "vectorized monstrosity using an 8-fold unrolled loop and a slew of SSE # instructions and XMM registers " # optimized assembly (as produced by Clang/LLVM 3.5.2). .text .file "/tmp/baz.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 factorial .align 16, 0x90 .type factorial,@function factorial: # @factorial .cfi_startproc # BB#0: movl $1, %eax testl %edi, %edi je .LBB0_26 # BB#1: # %overflow.checked movl %edi, %ecx andl $-8, %ecx movl %edi, %eax andl $-8, %eax je .LBB0_2 # BB#3: # %vector.body.preheader movl %edi, %r8d subl %ecx, %r8d leal -8(%rax), %edx shrl $3, %edx incl %edx je .LBB0_6 # BB#4: # %vector.body.preheader movl %edx, %ecx andl $1, %ecx jne .LBB0_6 # BB#5: pxor %xmm0, %xmm0 xorl %ecx, %ecx movdqa .LCPI0_0(%rip), %xmm2 pxor %xmm1, %xmm1 movdqa %xmm2, %xmm3 jmp .LBB0_7 .LBB0_2: xorl %eax, %eax movdqa .LCPI0_0(%rip), %xmm1 movl %edi, %r8d movdqa %xmm1, %xmm0 jmp .LBB0_11 .LBB0_6: # %vector.body.unr movd %edi, %xmm0 pshufd $0, %xmm0, %xmm0 # xmm0 = xmm0[0,0,0,0] movdqa .LCPI0_1(%rip), %xmm1 paddd %xmm0, %xmm1 paddd .LCPI0_2(%rip), %xmm0 movl $8, %ecx movdqa %xmm1, %xmm2 movdqa %xmm0, %xmm3 .LBB0_7: # %vector.body.preheader.split cmpl $2, %edx jb .LBB0_11 # BB#8: # %vector.body.preheader.split.split movl %edi, %edx subl %ecx, %edx movl %eax, %esi subl %ecx, %esi movdqa .LCPI0_1(%rip), %xmm8 movdqa .LCPI0_2(%rip), %xmm1 .align 16, 0x90 .LBB0_9: # %vector.body # =>This Inner Loop Header: Depth=1 movd %edx, %xmm4 pshufd $0, %xmm4, %xmm4 # xmm4 = xmm4[0,0,0,0] movdqa %xmm4, %xmm5 paddd %xmm8, %xmm5 paddd %xmm1, %xmm4 pshufd $49, %xmm5, %xmm6 # xmm6 = xmm5[1,0,3,0] pmuludq %xmm2, %xmm5 pshufd $49, %xmm2, %xmm2 # xmm2 = xmm2[1,0,3,0] pmuludq %xmm6, %xmm2 shufps $-120, %xmm2, %xmm5 # xmm5 = xmm5[0,2],xmm2[0,2] pshufd $-40, %xmm5, %xmm5 # xmm5 = xmm5[0,2,1,3] pshufd $49, %xmm4, %xmm6 # xmm6 = xmm4[1,0,3,0] pmuludq %xmm3, %xmm4 pshufd $49, %xmm3, %xmm3 # xmm3 = xmm3[1,0,3,0] pmuludq %xmm6, %xmm3 shufps $-120, %xmm3, %xmm4 # xmm4 = xmm4[0,2],xmm3[0,2] pshufd $-40, %xmm4, %xmm4 # xmm4 = xmm4[0,2,1,3] leal -8(%rdx), %ecx movd %ecx, %xmm6 pshufd $0, %xmm6, %xmm6 # xmm6 = xmm6[0,0,0,0] movdqa %xmm6, %xmm7 paddd %xmm8, %xmm7 paddd %xmm1, %xmm6 pshufd $49, %xmm7, %xmm0 # xmm0 = xmm7[1,0,3,0] pmuludq %xmm2, %xmm0 pmuludq %xmm5, %xmm7 shufps $-120, %xmm0, %xmm7 # xmm7 = xmm7[0,2],xmm0[0,2] pshufd $-40, %xmm7, %xmm2 # xmm2 = xmm7[0,2,1,3] pshufd $49, %xmm6, %xmm0 # xmm0 = xmm6[1,0,3,0] pmuludq %xmm3, %xmm0 pmuludq %xmm4, %xmm6 shufps $-120, %xmm0, %xmm6 # xmm6 = xmm6[0,2],xmm0[0,2] pshufd $-40, %xmm6, %xmm3 # xmm3 = xmm6[0,2,1,3] addl $-16, %edx addl $-16, %esi jne .LBB0_9 # BB#10: movdqa %xmm2, %xmm1 movdqa %xmm3, %xmm0 .LBB0_11: # %middle.block pshufd $49, %xmm0, %xmm2 # xmm2 = xmm0[1,0,3,0] pmuludq %xmm1, %xmm0 pshufd $49, %xmm1, %xmm1 # xmm1 = xmm1[1,0,3,0] pmuludq %xmm2, %xmm1 shufps $-120, %xmm1, %xmm0 # xmm0 = xmm0[0,2],xmm1[0,2] pshufd $-40, %xmm0, %xmm0 # xmm0 = xmm0[0,2,1,3] movdqa %xmm0, %xmm2 movhlps %xmm2, %xmm2 # xmm2 = xmm2[1,1] pmuludq %xmm0, %xmm2 movaps %xmm1, %xmm0 movhlps %xmm0, %xmm0 # xmm0 = xmm0[1,1] pmuludq %xmm1, %xmm0 shufps $-120, %xmm0, %xmm2 # xmm2 = xmm2[0,2],xmm0[0,2] pshufd $-40, %xmm2, %xmm1 # xmm1 = xmm2[0,2,1,3] pmuludq %xmm0, %xmm1 shufps $-120, %xmm1, %xmm1 # xmm1 = xmm1[0,2,0,2] pshufd $-40, %xmm1, %xmm0 # xmm0 = xmm1[0,2,1,3] movd %xmm0, %edx cmpl %edi, %eax jne .LBB0_13 # BB#12: movl %edx, %eax # kill: EAX EAX RAX retq .LBB0_13: # %tailrecurse.preheader movl %r8d, %ecx andl $3, %ecx testl %r8d, %r8d je .LBB0_16 # BB#14: # %tailrecurse.preheader xorl %eax, %eax testl %ecx, %ecx jne .LBB0_16 # BB#15: movl %r8d, %esi jmp .LBB0_23 .LBB0_16: # %unr.cmp15 cmpl $1, %ecx jne .LBB0_18 # BB#17: movl %r8d, %eax jmp .LBB0_22 .LBB0_18: # %unr.cmp cmpl $2, %ecx jne .LBB0_20 # BB#19: movl %r8d, %ecx jmp .LBB0_21 .LBB0_20: # %tailrecurse.unr leal -1(%r8), %ecx imull %r8d, %edx .LBB0_21: # %tailrecurse.unr11 leal -1(%rcx), %eax imull %edx, %ecx movl %ecx, %edx .LBB0_22: # %tailrecurse.unr12 leal -1(%rax), %esi imull %edx, %eax movl %eax, %edx .LBB0_23: # %tailrecurse.preheader.split cmpl $4, %r8d jb .LBB0_26 # BB#24: movl %edx, %eax .align 16, 0x90 .LBB0_25: # %tailrecurse # =>This Inner Loop Header: Depth=1 leal -1(%rsi), %ecx imull %esi, %eax imull %ecx, %eax leal -2(%rsi), %ecx imull %eax, %ecx leal -3(%rsi), %eax imull %ecx, %eax addl $-4, %esi jne .LBB0_25 .LBB0_26: # %tailrecurse._crit_edge # kill: EAX EAX RAX retq .Ltmp0: .size factorial, .Ltmp0-factorial .cfi_endproc .ident "Debian clang version 3.5.2-2 (tags/RELEASE_352/final) (based on LLVM 3.5.2)" .section ".note.GNU-stack","",@progbits