simple loop: vectorisation with gcc

On my laptop (ubuntu.14 + gcc-5.x), I have AVX:

~> tail /proc/cpuinfo 
   model name   : Intel(R) Core(TM) i7-3687U CPU @ 2.10GHz
   flags        : ... sse sse2 ... avx

I compile this very simple code:

~> more test.c 
  #include <stdio.h>
  void main() {
    int i=0; double a=1.;
    for(i=0;i<1000000;i++) a+=i;
    printf("%f\n", a); // printf: avoid compiler optim (dummy var suppression)
  }
~> make
   gcc -O2 -march=native -mavx -ftree-vectorize -funroll-loops -fopt-info-vec -fopt-info-loop -o test.exe test.c
   test.c:4:3: note: loop unrolled 7 times

I don't understand if the loop as been "really" vectorised as the message says ! Objdump tells:

~> objdump -S test.exe | grep add
   40041d:  48 83 c4 08             add    $0x8,%rsp
   4004a9:  c5 fb 58 c2             vaddsd %xmm2,%xmm0,%xmm0
   4004b8:  c5 fb 58 ec             vaddsd %xmm4,%xmm0,%xmm5
   4004cb:  c5 53 58 c7             vaddsd %xmm7,%xmm5,%xmm8
   4004e1:  c4 41 3b 58 da          vaddsd %xmm10,%xmm8,%xmm11
   4004ea:  83 c0 08                add    $0x8,%eax
   4004f2:  c4 41 23 58 f5          vaddsd %xmm13,%xmm11,%xmm14
   4004f7:  c5 8b 58 d1             vaddsd %xmm1,%xmm14,%xmm2
   4004fb:  c5 eb 58 e3             vaddsd %xmm3,%xmm2,%xmm4
   4004ff:  c5 db 58 c6             vaddsd %xmm6,%xmm4,%xmm0
   4005bb:  48 01 c6                add    %rax,%rsi
   40067d:  48 83 c3 01             add    $0x1,%rbx
   400686:  48 83 c4 08             add    $0x8,%rsp
   4006a8:  48 83 c4 08             add    $0x8,%rsp

So finally, I get "vaddsd" (whit a "v" that seems to stand for "vectorised") but I do not have the "addpd" I would have expected ?...

My understanding is that "addsd" is scalar addition (= 1 regular addition), and that "addpd" is packed addition (= several additions vectorised in 1 cycle). Also, I don't understand in what "vaddsd" is different from "addpd" : are these supposed to be the same ? (google this does not give relevant answers)

Why don't I get "addpd" ? Missing compile option ? Missing hints / pragma in the code ? Or is it logical, if yes why ?

FH

UPDATE

The message says it has been vectorized but it's not, I get no speed-up :

~> more test.c 
   #include <stdio.h>
   void main() {
     unsigned int i=0; double a=1.;
     for(i=0;i<3000000000;i++) a+=i;
     printf("%f\n", a); // printf : avoid compiler optimisation what suppress a as it's a dummy variable !
   }
~> make
   gcc -O2 -march=native -o test.novec.exe test.c
   gcc -O2 -march=native -mavx -ftree-vectorize -funroll-loops -fopt-info-vec -fopt-info-loop -o test.vec.exe test.c
   test.c:4:3: note: loop unrolled 7 times
~> time ./test.novec.exe 
   4499999997067113984.000000
   real 0m2.927s
   user 0m2.928s
   sys  0m0.000s
~> time ./test.vec.exe
   4499999997067113984.000000
   real 0m2.926s
   user 0m2.924s
   sys  0m0.000s

... Unless I add -ffast-math (or -Ofast that includes -ffast-math) :

~> make
   gcc -O2 -march=native -mavx -ftree-vectorize -funroll-loops -fopt-info-vec -fopt-info-loop -ffast-math -o test.vec.fm.exe test.c
   test.c:4:3: note: loop vectorized
   test.c:4:30: note: loop unrolled 3 times
~> time ./test.vec.fm.exe 
   4499999999597346816.000000
   real 0m1.980s
   user 0m1.980s
   sys  0m0.000s