auto-vectorisation example in GCC

gistfile1.txt

void test_vec(double* restrict a, double* restrict b) {
    int i;
 
    double *x = __builtin_assume_aligned(a, 16);
    double *y = __builtin_assume_aligned(b, 16);
 
    for(i = 0; i < SIZE; i++) {
        x[i] = x[i] + y[i];
    }
}
 
-------------------------------------------------------------------------------------------
 
$ c99 -S -O2 autovec.c -o output.s (GNU Compiler Collection)
 
test_vec
.LFB22:
    .cfi_startproc
    xorl    %eax, %eax
    .p2align 4,,10
    .p2align 3
.L3:
    movsd   (%rdi,%rax), %xmm0
    addsd   (%rsi,%rax), %xmm0
    movsd   %xmm0, (%rdi,%rax)
    addq    $8, %rax
    cmpq    $256, %rax
    jne .L3 
    rep ret 
    .cfi_endproc

-------------------------------------------------------------------------------------------
 
$ c99 -S -O3 autovec.c -o output.s (GNU Compiler Collection)
 
  4006d0:	31 c0                	xor    %eax,%eax
  4006d2:	66 0f 1f 44 00 00    	nopw   0x0(%rax,%rax,1)
  4006d8:	66 0f 28 04 07       	movapd (%rdi,%rax,1),%xmm0
  4006dd:	66 0f 58 04 06       	addpd  (%rsi,%rax,1),%xmm0
  4006e2:	66 0f 29 04 07       	movapd %xmm0,(%rdi,%rax,1)
  4006e7:	48 83 c0 10          	add    $0x10,%rax
  4006eb:	48 3d 00 00 08 00    	cmp    $0x80000,%rax
  4006f1:	75 e5                	jne    4006d8 <test_vec+0x8>
  4006f3:	f3 c3                	repz retq 
  4006f5:	66 66 2e 0f 1f 84 00 	data32 nopw %cs:0x0(%rax,%rax,1)

http://locklessinc.com/articles/vectorize/