• Locked Locked
  • Replies 2 replies
  • Subscribers 119 subscribers
  • Views 3645 views
  • Users 0 members are here
2025 survey

We are running a survey to help us improve the experience for all of our members. If you see the survey appear, please take the time to tell us about your experience if you can.

Options
Related
How was your experience today?
This discussion has been locked.
You can no longer post new replies to this discussion. If you have a question you can start a new discussion

Optimising with NEON

Roumen Popov
Note: This was originally posted on 11th February 2010 at http://forums.arm.com

Hello guys, I am a bit new with ARM and NEON and was wondering if anyone could help me with advice on how to optimise a small piece of code by NEON.
The problem is that whatever I do I can't get the NEON code to execute faster than the standard C code.
I am running the code on Beagle board with the latest Angstrom distribution and building with the Code Sourcery Lite gcc.
My gcc command line options are  -O3 -mcpu=cortex-a8 -mfpu=neon -mfloat-abi=softfp -ffast-math
the piece of code in question is the following:

int32_t weightsH[4];
int16_t weightsV[4];
int16_t *lut1, *lut2;

// start of loop

quantLevel1 = ...
quantLevel2 = ...
frac = ...
frac_1 = ...

pxlOut      = 0;

lut1 = luts[quantLevel1];
lut2 = luts[quantLevel2];

w1 = lut1[0];
w2 = lut2[0];
weight = ((weightsH[0]>>16)*weightsV[0])>>13;
pxlOut += (weight*((w1*frac + w2*frac_1)>>14))>>14;

w1 = lut1[1];
w2 = lut2[1];
weight = ((weightsH[1]>>16)*weightsV[0])>>13;
pxlOut += (weight*((w1*frac + w2*frac_1)>>14))>>14;

w1 = lut1[2];
w2 = lut2[2];
weight = ((weightsH[0]>>16)*weightsV[1])>>13;
pxlOut += (weight*((w1*frac + w2*frac_1)>>14))>>14;

w1 = lut1[3];
w2 = lut2[3];
weight = ((weightsH[1]>>16)*weightsV[1])>>13;
pxlOut += (weight*((w1*frac + w2*frac_1)>>14))>>14;

weightsH[0] -= slope_x;
weightsH[1] += slope_x;

// end of loop

the NEON code with the best results that I managed to obtain is the following, yet it's still slower than the above, it's just killing me:

int16x4_t wVs, lt1, lt2, fr1, w;
int32x4_t wHs,s;
int32x4_t buf;
int32x2_t buf2;

slopes[0] = -slope_x;
slopes[1] = slope_x;
slopes[2] = -slope_x;
slopes[3] = slope_x;

wHs = vld1q_s32(weightsH);
s = vld1q_s32(slopes);
wVs = vld1_s16(weightsV);

// start of loop

quantLevel1 = ...
quantLevel2 = ...
frac = ...
frac_1 = ...

lut1 = luts[quantLevel1];
lut2 = luts[quantLevel2];

buf = vdupq_n_s32(0);
lt1 = vld1_s16(lut1);
lt2 = vld1_s16(lut2);
buf = vmlal_n_s16(buf, lt1, (int16_t)frac);
buf = vmlal_n_s16(buf, lt2, (int16_t)frac_1);
fr1 = vshrn_n_s32(buf, 14);
w = vshrn_n_s32(wHs, 16);
buf = vmull_s16(w, wVs);
w = vshrn_n_s32(buf, 13);
buf = vmull_s16(w, fr1);
fr1 = vshrn_n_s32(buf, 14);
buf2 = vpaddl_s16(fr1);
pxlOut = vget_lane_s32(buf2, 0) + vget_lane_s32(buf2, 1);

// end of loop

Thank you in advance,
Roumen
Parents
  • Arm Employee Badge
    Note: This was originally posted on 11th February 2010 at http://forums.arm.com

    Roumen,

    Have you looked at what code the compiler is generating for the pure C version,
    with the command line options you're using, it's likely that GCC is already using Neon,
    e.g.

    typedef int t;

    t foo(t *p)
    {
    int i;
    t max;
    max = p[0];

    for(i=0;i<32;i++)
      if(p[i]>max) max=p[i];

    return max;
    }

    Compiled with GCC 4.4.1 using "-O3 -mcpu=cortex-a8 -mfpu=neon -mfloat-abi=softfp -ffast-math" automatically produces the Neon code:

    mov  r1, r0
    ldr  r2, [r0, #0]
    vld1.32 {d2}, [r1]!
    sub  sp, sp, #8
    vld1.32 {d20}, [r1]!
    vld1.32 {d0}, [r1]!
    str  r2, [sp, #4]
    vld1.32 {d22}, [r1]!
    str  r2, [sp, #0]
    fldd d16, [sp, #0]
    vmax.s32  d1, d2, d16
    vmax.s32  d17, d1, d20
    vld1.32 {d19}, [r1]!
    vmax.s32  d31, d17, d0
    vld1.32 {d21}, [r1]!
    vmax.s32  d30, d31, d22
    vld1.32 {d27}, [r1]!
    vmax.s32  d29, d30, d19
    vld1.32 {d20}, [r1]!
    vmax.s32  d28, d29, d21
    vld1.32 {d18}, [r1]!
    vmax.s32  d26, d28, d27
    vld1.32 {d19}, [r1]!
    vmax.s32  d25, d26, d20
    vld1.32 {d7}, [r1]!
    vmax.s32  d24, d25, d18
    vld1.32 {d18}, [r1]!
    vmax.s32  d23, d24, d19
    vld1.32 {d4}, [r1]!
    vmax.s32  d6, d23, d7
    vld1.32 {d2}, [r1]!
    vmax.s32  d5, d6, d18
    vld1.32 {d1}, [r1]
    vmax.s32  d3, d5, d4
    add  r3, r1, #8
    vmax.s32  d17, d3, d2
    vmax.s32  d0, d17, d1
    vld1.32 {d17}, [r3]
    vmax.s32  d16, d17, d0
    vpmax.s32    d16, d16, d16
    vmov.32 r0, d16[0]
    add  sp, sp, #8
    bx   lr

    hth
    s.
Reply
  • Arm Employee Badge
    Note: This was originally posted on 11th February 2010 at http://forums.arm.com

    Roumen,

    Have you looked at what code the compiler is generating for the pure C version,
    with the command line options you're using, it's likely that GCC is already using Neon,
    e.g.

    typedef int t;

    t foo(t *p)
    {
    int i;
    t max;
    max = p[0];

    for(i=0;i<32;i++)
      if(p[i]>max) max=p[i];

    return max;
    }

    Compiled with GCC 4.4.1 using "-O3 -mcpu=cortex-a8 -mfpu=neon -mfloat-abi=softfp -ffast-math" automatically produces the Neon code:

    mov  r1, r0
    ldr  r2, [r0, #0]
    vld1.32 {d2}, [r1]!
    sub  sp, sp, #8
    vld1.32 {d20}, [r1]!
    vld1.32 {d0}, [r1]!
    str  r2, [sp, #4]
    vld1.32 {d22}, [r1]!
    str  r2, [sp, #0]
    fldd d16, [sp, #0]
    vmax.s32  d1, d2, d16
    vmax.s32  d17, d1, d20
    vld1.32 {d19}, [r1]!
    vmax.s32  d31, d17, d0
    vld1.32 {d21}, [r1]!
    vmax.s32  d30, d31, d22
    vld1.32 {d27}, [r1]!
    vmax.s32  d29, d30, d19
    vld1.32 {d20}, [r1]!
    vmax.s32  d28, d29, d21
    vld1.32 {d18}, [r1]!
    vmax.s32  d26, d28, d27
    vld1.32 {d19}, [r1]!
    vmax.s32  d25, d26, d20
    vld1.32 {d7}, [r1]!
    vmax.s32  d24, d25, d18
    vld1.32 {d18}, [r1]!
    vmax.s32  d23, d24, d19
    vld1.32 {d4}, [r1]!
    vmax.s32  d6, d23, d7
    vld1.32 {d2}, [r1]!
    vmax.s32  d5, d6, d18
    vld1.32 {d1}, [r1]
    vmax.s32  d3, d5, d4
    add  r3, r1, #8
    vmax.s32  d17, d3, d2
    vmax.s32  d0, d17, d1
    vld1.32 {d17}, [r3]
    vmax.s32  d16, d17, d0
    vpmax.s32    d16, d16, d16
    vmov.32 r0, d16[0]
    add  sp, sp, #8
    bx   lr

    hth
    s.
Children
No data