FWIW, __int128 and the optimizations mentioned works in clang too, my guess is that the author just runs an old version. #include <stdio.h> int ma

FWIW, __int128 and the optimizations mentioned works in clang too, my guess is that the author just runs an old version. #include <stdio.h> int main() { FILE *fp; __int128 v=1234; fp = fopen("foo12", "w"); fwrite(&v, sizeof(v), 1, fp); fclose(fp); return 0; } with -O2 -march=native -ffinite-math-only -fno-math-errno, for clang 3.5.0, linux/amd64, gives me: 4005d0: 53 push %rbx 4005d1: 48 83 ec 10 sub $0x10,%rsp 4005d5: 48 c7 44 24 08 00 00 movq $0x0,0x8(%rsp) 4005dc: 00 00 4005de: 48 c7 04 24 d2 04 00 movq $0x4d2,(%rsp) 4005e5: 00 4005e6: bf b0 06 40 00 mov $0x4006b0,%edi 4005eb: be b6 06 40 00 mov $0x4006b6,%esi 4005f0: e8 bb fe ff ff callq 4004b0 <fopen@plt> 4005f5: 48 89 c3 mov %rax,%rbx 4005f8: 48 8d 3c 24 lea (%rsp),%rdi 4005fc: be 10 00 00 00 mov $0x10,%esi 400601: ba 01 00 00 00 mov $0x1,%edx 400606: 48 89 d9 mov %rbx,%rcx 400609: e8 b2 fe ff ff callq 4004c0 <fwrite@plt> 40060e: 48 89 df mov %rbx,%rdi 400611: e8 6a fe ff ff callq 400480 <fclose@plt> 400616: 31 c0 xor %eax,%eax 400618: 48 83 c4 10 add $0x10,%rsp 40061c: 5b pop %rbx 40061d: c3 retq 40061e: 66 90 xchg %ax,%ax