Printing Armstrong numbers from 1 to 500. (C Programming)

pan64 · 06-14-2017, 05:42 AM

Quote:

Originally Posted by Laserbeak

OMG, I actually did it with SSE!!! This is my first vector program (with some help from reading a lot of things and one post to another site):

It'd be great if you can test its speed!

Sorry guy, how should I compile/build it?

Laserbeak · 06-14-2017, 11:07 AM

Quote:

Originally Posted by pan64

Sorry guy, how should I compile/build it?

As long as you have an advanced enough processor and compiler, it should just build. Have you tried and gotten an error? If so, please post it.

pan64 · 06-14-2017, 11:23 AM

Code:

user@host:/tmp/armstrong$ gcc -msse4.1 -o a a.c
a.c: In function ‘main’:
a.c:20:14: error: incompatible type for argument 1 of ‘_mm_hadd_ps’
         v3 = _mm_hadd_ps(v3,v3);
              ^
In file included from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/tmmintrin.h:31:0,
                 from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/smmintrin.h:32,
                 from a.c:2:
/usr/lib/gcc/x86_64-linux-gnu/4.9/include/pmmintrin.h:56:1: note: expected ‘__m128’ but argument is of type ‘__m128i’
 _mm_hadd_ps (__m128 __X, __m128 __Y)
 ^
a.c:20:14: error: incompatible type for argument 2 of ‘_mm_hadd_ps’
         v3 = _mm_hadd_ps(v3,v3);
              ^
In file included from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/tmmintrin.h:31:0,
                 from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/smmintrin.h:32,
                 from a.c:2:
/usr/lib/gcc/x86_64-linux-gnu/4.9/include/pmmintrin.h:56:1: note: expected ‘__m128’ but argument is of type ‘__m128i’
 _mm_hadd_ps (__m128 __X, __m128 __Y)
 ^
a.c:21:14: error: incompatible type for argument 1 of ‘_mm_hadd_ps’
         v3 = _mm_hadd_ps(v3,v3);
              ^
In file included from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/tmmintrin.h:31:0,
                 from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/smmintrin.h:32,
                 from a.c:2:
/usr/lib/gcc/x86_64-linux-gnu/4.9/include/pmmintrin.h:56:1: note: expected ‘__m128’ but argument is of type ‘__m128i’
 _mm_hadd_ps (__m128 __X, __m128 __Y)
 ^
a.c:21:14: error: incompatible type for argument 2 of ‘_mm_hadd_ps’
         v3 = _mm_hadd_ps(v3,v3);
              ^
In file included from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/tmmintrin.h:31:0,
                 from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/smmintrin.h:32,
                 from a.c:2:
/usr/lib/gcc/x86_64-linux-gnu/4.9/include/pmmintrin.h:56:1: note: expected ‘__m128’ but argument is of type ‘__m128i’
 _mm_hadd_ps (__m128 __X, __m128 __Y)
 ^
user@host:/tmp/armstrong$ gcc --version
gcc (Debian 4.9.2-10) 4.9.2

AMD FX-8320

Laserbeak · 06-14-2017, 11:25 AM

I looped over it 1000x from within C (not rerunning the program), not printing the results (I did print the results once and they were all correct) and got:

Code:

real	0m0.005s
user	0m0.001s
sys	0m0.001s

10000x:

Code:

real	0m0.005s
user	0m0.001s
sys	0m0.002s

100000x:

Code:

real	0m0.005s
user	0m0.001s
sys	0m0.001s

1000000x:

Code:

real	0m0.004s
user	0m0.001s
sys	0m0.001s

So basically the documentation is true, SSE commands are basically NOOPs.

This the 1,000,000 time code:

Code:

#include <stdio.h>
#include <smmintrin.h>  // SSE 4.1

__m128i vcube(const __m128i v)
{
    return _mm_mullo_epi32(v, _mm_mullo_epi32(v, v));
}


int main(int argc, const char * argv[]) {
    for (unsigned int y = 0; y < 1000000; y++) {
        for (unsigned int i = 1; i <= 500; i++) {
            unsigned int firstDigit = i / 100;
            unsigned int secondDigit = (i - firstDigit * 100) / 10;
            unsigned int thirdDigit = (i - firstDigit * 100 - secondDigit * 10);
            
            __m128i v = _mm_setr_epi32(0, firstDigit, secondDigit, thirdDigit);
            __m128i v3 = vcube(v);
            
            v3 = _mm_hadd_ps(v3,v3);
            v3 = _mm_hadd_ps(v3,v3);

       /*     if (_mm_extract_epi32(v3, 0) == i)
                printf ("%d is an Armstrong number\n", i);
        */
        }
    }
    return 0;
}

Laserbeak · 06-14-2017, 11:30 AM

Quote:

Originally Posted by pan64

Code:

user@host:/tmp/armstrong$ gcc -msse4.1 -o a a.c
a.c: In function ‘main’:
a.c:20:14: error: incompatible type for argument 1 of ‘_mm_hadd_ps’
         v3 = _mm_hadd_ps(v3,v3);
              ^
In file included from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/tmmintrin.h:31:0,
                 from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/smmintrin.h:32,
                 from a.c:2:
/usr/lib/gcc/x86_64-linux-gnu/4.9/include/pmmintrin.h:56:1: note: expected ‘__m128’ but argument is of type ‘__m128i’
 _mm_hadd_ps (__m128 __X, __m128 __Y)
 ^
a.c:20:14: error: incompatible type for argument 2 of ‘_mm_hadd_ps’
         v3 = _mm_hadd_ps(v3,v3);
              ^
In file included from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/tmmintrin.h:31:0,
                 from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/smmintrin.h:32,
                 from a.c:2:
/usr/lib/gcc/x86_64-linux-gnu/4.9/include/pmmintrin.h:56:1: note: expected ‘__m128’ but argument is of type ‘__m128i’
 _mm_hadd_ps (__m128 __X, __m128 __Y)
 ^
a.c:21:14: error: incompatible type for argument 1 of ‘_mm_hadd_ps’
         v3 = _mm_hadd_ps(v3,v3);
              ^
In file included from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/tmmintrin.h:31:0,
                 from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/smmintrin.h:32,
                 from a.c:2:
/usr/lib/gcc/x86_64-linux-gnu/4.9/include/pmmintrin.h:56:1: note: expected ‘__m128’ but argument is of type ‘__m128i’
 _mm_hadd_ps (__m128 __X, __m128 __Y)
 ^
a.c:21:14: error: incompatible type for argument 2 of ‘_mm_hadd_ps’
         v3 = _mm_hadd_ps(v3,v3);
              ^
In file included from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/tmmintrin.h:31:0,
                 from /usr/lib/gcc/x86_64-linux-gnu/4.9/include/smmintrin.h:32,
                 from a.c:2:
/usr/lib/gcc/x86_64-linux-gnu/4.9/include/pmmintrin.h:56:1: note: expected ‘__m128’ but argument is of type ‘__m128i’
 _mm_hadd_ps (__m128 __X, __m128 __Y)
 ^
user@host:/tmp/armstrong$ gcc --version
gcc (Debian 4.9.2-10) 4.9.2

AMD FX-8320

I don't really know... the one thing that sticks out is you have AMD and I know Intel's implementation is slightly different. Like I said this was my first attempt at SSE instructions so maybe someone else can help. But it certainly compiles file on my Mac:

Code:

________________________________________________________________________________
| ~/Library/Mobile Documents/com~apple~CloudDocs/AppDevel/armstrongsse/armstrongsse @ SpocksBrain (djohnsto) 
| => cat main.c
//
//  main.c
//  armstrongsse
//
//  Created by Douglas Johnston on 6/13/17.
//  Copyright © 2017 Douglas Johnston. All rights reserved.
//

#include <stdio.h>
#include <smmintrin.h>  // SSE 4.1

__m128i vcube(const __m128i v)
{
    return _mm_mullo_epi32(v, _mm_mullo_epi32(v, v));
}


int main(int argc, const char * argv[]) {
    for (unsigned int y = 0; y < 1000000; y++) {
        for (unsigned int i = 1; i <= 500; i++) {
            unsigned int firstDigit = i / 100;
            unsigned int secondDigit = (i - firstDigit * 100) / 10;
            unsigned int thirdDigit = (i - firstDigit * 100 - secondDigit * 10);
            
            __m128i v = _mm_setr_epi32(0, firstDigit, secondDigit, thirdDigit);
            __m128i v3 = vcube(v);
            
            v3 = _mm_hadd_ps(v3,v3);
            v3 = _mm_hadd_ps(v3,v3);

       /*     if (_mm_extract_epi32(v3, 0) == i)
                printf ("%d is an Armstrong number\n", i);
        */
        }
    }
    return 0;
}
________________________________________________________________________________
| ~/Library/Mobile Documents/com~apple~CloudDocs/AppDevel/armstrongsse/armstrongsse @ SpocksBrain (djohnsto) 
| => clang -O3 -o armtestsse main.c
________________________________________________________________________________
| ~/Library/Mobile Documents/com~apple~CloudDocs/AppDevel/armstrongsse/armstrongsse @ SpocksBrain (djohnsto) 
| => time ./armtestsse

real	0m0.005s
user	0m0.001s
sys	0m0.001s
________________________________________________________________________________
| ~/Library/Mobile Documents/com~apple~CloudDocs/AppDevel/armstrongsse/armstrongsse @ SpocksBrain (djohnsto) 
| =>

Forget that copyright s***, Xcode adds that automatically.

Laserbeak · 06-14-2017, 11:45 AM

I changed it to still test, but got the really the same numbers:

Code:

#include <stdio.h>
#include <smmintrin.h>  // SSE 4.1
#include <stdbool.h>

__m128i vcube(const __m128i v)
{
    return _mm_mullo_epi32(v, _mm_mullo_epi32(v, v));
}


int main(int argc, const char * argv[]) {
    for (unsigned int y = 0; y < 1000000; y++) {
        for (unsigned int i = 1; i <= 500; i++) {
            bool isArmstrong = false;
            unsigned int firstDigit = i / 100;
            unsigned int secondDigit = (i - firstDigit * 100) / 10;
            unsigned int thirdDigit = (i - firstDigit * 100 - secondDigit * 10);
            
            __m128i v = _mm_setr_epi32(0, firstDigit, secondDigit, thirdDigit);
            __m128i v3 = vcube(v);
            
            v3 = _mm_hadd_ps(v3,v3);
            v3 = _mm_hadd_ps(v3,v3);

            if (_mm_extract_epi32(v3, 0) == i)
        
                isArmstrong = true;
        }
    }
    return 0;
}

pan64 · 06-14-2017, 11:59 AM

if I remember well it happened on an intel based laptop too, but i will try to check it tomorrow. From the other hand this CPU supports SSE4.1.

Code:

flags		: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc extd_apicid aperfmperf pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 popcnt aes xsave avx f16c lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs xop skinit wdt lwp fma4 tce nodeid_msr tbm topoext perfctr_core perfctr_nb arat cpb hw_pstate npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold vmmcall bmi1

it looks like type mismatch/incompatibility:
_mm_hadd_ps returns __m128, not __m128i.

Laserbeak · 06-14-2017, 12:22 PM

Let me try it in Solaris... rebooting...

Laserbeak · 06-14-2017, 12:57 PM

I ran into some compilation problems in Solaris too. I was able to fix them and this printed out the correct answers... hopefully it'll work on your compiler:

Code:

#include <stdio.h>
#include <smmintrin.h>  // SSE 4.1

__m128i vcube(const __m128i v)
{
    return _mm_mullo_epi32(v, _mm_mullo_epi32(v, v));
}


int main(int argc, const char * argv[]) {
    for (unsigned int i = 1; i <= 500; i++) {
        unsigned int firstDigit = i / 100;
        unsigned int secondDigit = (i - firstDigit * 100) / 10;
        unsigned int thirdDigit = (i - firstDigit * 100 - secondDigit * 10);

        __m128i v = _mm_setr_epi32(0, firstDigit, secondDigit, thirdDigit);
        __m128 v3 = (__m128) vcube(v);

        v3 = _mm_hadd_ps(v3, v3);
        v3 = _mm_hadd_ps(v3, v3);

        if (_mm_extract_epi32((__m128i) v3, 0) == i)
            printf ("%d is an Armstrong number\n", i);
    }
    return 0;
}

I used this to compile:

Code:

gcc -std=c99 -m64 -msse4.1 -O3 -o testsse testsse.c

Also, the same code works unchanged on the Mac. It seemingly is just more forgiving when it comes to typing.

pan64 · 06-14-2017, 01:14 PM

Code:

            normal   sse4.1
compiled   0.85    14.91
gcc -O3    0.31     9.00

if I did not miss anything.

Laserbeak · 06-14-2017, 01:23 PM

Quote:

Originally Posted by pan64

Code:

            normal   sse4.1
compiled   0.85    14.91
gcc -O3    0.31     9.00

if I did not miss anything.

So you're saying it's a lot SLOWER on your machine? Hmm... can anyone else confirm that? I get much faster rates.

What type of processor do you have? Maybe it's emulating these instructions in software, not executing them directly in hardware. My computer shows them to be much faster...

I don't mean to make this a d**k measuring contest, but it's definitely interesting why you seemingly got so different speed measurements.

pan64 · 06-14-2017, 01:34 PM

Yes, it looks like. I have no idea why. I have AMD FX-8320, it supports SSE4.1.
I have just tried to debug this and it looks like it used the real SSE, there was no emulation. But again, probably I missed something.

Laserbeak · 06-14-2017, 01:38 PM

Quote:

Originally Posted by pan64

Yes, it looks like. I have no idea why. I have AMD FX-8320, it supports SSE4.1.
I have just tried to debug this and it looks like it used the real SSE, there was no emulation. But again, probably I missed something.

OK, well I have the Intel Core i7 2.3 GHz that came with the MacBook Pro (Retina, 15-inch, Late 2013). Maybe someone who knows more about this can shed some light on it.

KenJackson · 06-14-2017, 01:42 PM

Quote:

Originally Posted by Laserbeak

Let me try it in Solaris... rebooting...

Real Solaris? Or one of the illumos distros?

Laserbeak · 06-14-2017, 01:54 PM

Quote:

Originally Posted by KenJackson

Real Solaris? Or one of the illumos distros?

Real Solaris x86_64 11.3