Demo of vectorizing sqrt using LLVM 3.5.0

gistfile1.txt

julia> function f(x,y,z)
           @fastmath @inbounds @simd for i=1:length(z)
               z[i] = sqrt(x[i]^2+y[i]^2)
           end
       end
f (generic function with 1 method)

julia> code_native(f,(Vector{Float32},Vector{Float32},Vector{Float32}))
        .text
Filename: none
Source line: 0
        push    rbp
        mov     rbp, rsp
Source line: 49
        mov     rax, qword ptr [rdx + 16]
        xor     r11d, r11d
        test    rax, rax
        cmovns  r11, rax
        dec     r11
        jo      L343
        inc     r11
        jo      L343
Source line: 50
        test    r11, r11
        jle     L333
Source line: 3
        mov     r8, qword ptr [rdi + 8]
        mov     r9, qword ptr [rsi + 8]
        mov     r10, qword ptr [rdx + 8]
        xor     esi, esi
        test    r11, r11
        je      L263
        xor     esi, esi
Source line: 56
        mov     rdi, r11
        and     rdi, -32
        je      L258
        lea     rcx, qword ptr [r10 + 96]
        lea     rsi, qword ptr [r9 + 96]
        lea     rdx, qword ptr [r8 + 96]
        mov     rax, r11
        and     rax, -32
        nop     word ptr cs:[rax + rax]
Source line: 3
L112:   vmovups ymm0, ymmword ptr [rdx - 96]
        vmovups ymm1, ymmword ptr [rdx - 64]
        vmovups ymm2, ymmword ptr [rdx - 32]
        vmovups ymm3, ymmword ptr [rdx]
        vmulps  ymm0, ymm0, ymm0
        vmulps  ymm1, ymm1, ymm1
        vmulps  ymm2, ymm2, ymm2
        vmulps  ymm3, ymm3, ymm3
        vmovups ymm4, ymmword ptr [rsi - 96]
        vmovups ymm5, ymmword ptr [rsi - 64]
        vmovups ymm6, ymmword ptr [rsi - 32]
        vmovups ymm7, ymmword ptr [rsi]
        vmulps  ymm4, ymm4, ymm4
        vmulps  ymm5, ymm5, ymm5
        vmulps  ymm6, ymm6, ymm6
        vmulps  ymm7, ymm7, ymm7
        vaddps  ymm0, ymm0, ymm4
        vaddps  ymm1, ymm1, ymm5
        vaddps  ymm2, ymm2, ymm6
        vaddps  ymm3, ymm3, ymm7
        vsqrtps ymm0, ymm0
        vsqrtps ymm1, ymm1
        vsqrtps ymm2, ymm2
        vsqrtps ymm3, ymm3
        vmovups ymmword ptr [rcx - 96], ymm0
        vmovups ymmword ptr [rcx - 64], ymm1
        vmovups ymmword ptr [rcx - 32], ymm2
        vmovups ymmword ptr [rcx], ymm3
Source line: 56
        sub     rcx, -128
        sub     rsi, -128
        sub     rdx, -128
        add     rax, -32
        jne     L112
        mov     rsi, rdi
L258:   cmp     r11, rsi
        je      L333
L263:   lea     rax, qword ptr [r10 + 4*rsi]
        lea     rcx, qword ptr [r9 + 4*rsi]
        lea     rdx, qword ptr [r8 + 4*rsi]
        sub     r11, rsi
        nop     word ptr cs:[rax + rax]
Source line: 3
L288:   vmovss  xmm0, dword ptr [rdx]
        vmulss  xmm0, xmm0, xmm0
        vmovss  xmm1, dword ptr [rcx]
        vmulss  xmm1, xmm1, xmm1
        vaddss  xmm0, xmm0, xmm1
        vsqrtss xmm0, xmm0, xmm0
        vmovss  dword ptr [rax], xmm0
Source line: 57
        add     rax, 4
        add     rcx, 4
        add     rdx, 4
        dec     r11
        jne     L288
L333:   mov     eax, 12394448
Source line: 64
        pop     rbp
        vzeroupper
        ret
Source line: 49
L343:   movabs  rax, jl_overflow_exception
        mov     rdi, qword ptr [rax]
        movabs  rax, jl_throw_with_superfluous_argument
        mov     esi, 49
        call    rax

julia>