diff options
| -rw-r--r-- | src/vnet/ip/ip_packet.h | 16 |
1 files changed, 15 insertions, 1 deletions
diff --git a/src/vnet/ip/ip_packet.h b/src/vnet/ip/ip_packet.h index b0b5f41260c..d862caa3a52 100644 --- a/src/vnet/ip/ip_packet.h +++ b/src/vnet/ip/ip_packet.h @@ -301,6 +301,20 @@ always_inline u16 ip_csum_fold (ip_csum_t c) { /* Reduce to 16 bits. */ +#ifdef __x86_64__ + u64 tmp; + asm volatile( + /* using ADC is much faster than mov, shift, add sequence + * compiler produces */ + "mov %k[sum], %k[tmp] \n\t" + "shr $32, %[sum] \n\t" + "add %k[tmp], %k[sum] \n\t" + "mov $16, %k[tmp] \n\t" + "shrx %k[tmp], %k[sum], %k[tmp] \n\t" + "adc %w[tmp], %w[sum] \n\t" + "adc $0, %w[sum] \n\t" + : [ sum ] "+&r"(c), [ tmp ] "=&r"(tmp)); +#else #if uword_bits == 64 c = (c & (ip_csum_t) 0xffffffff) + (c >> (ip_csum_t) 32); c = (c & 0xffff) + (c >> 16); @@ -308,7 +322,7 @@ ip_csum_fold (ip_csum_t c) c = (c & 0xffff) + (c >> 16); c = (c & 0xffff) + (c >> 16); - +#endif return c; } |