And the guess how it's arriving at that optimization:
Step 0: base code
x := 2.5
res := 0.0
for n := 0; n < NMAX; n++ {
for i := 0; i < 4; i++ {
res += math.Pow(x, float(i))
}
}
Step 1: unroll inner-loop
x := 2.5
res := 0.0
for n := 0; n < NMAX; n++ {
res += math.Pow(x, float(0))
res += math.Pow(x, float(1))
res += math.Pow(x, float(2))
res += math.Pow(x, float(3))
}
Step 2: detect that x doesn't change and inline
res := 0.0
for n := 0; n < NMAX; n++ {
res += math.Pow(2.5, float(0))
res += math.Pow(2.5, float(1))
res += math.Pow(2.5, float(2))
res += math.Pow(2.5, float(3))
}
Step 3: compute constant function Pow
res := 0.0
for n := 0; n < NMAX; n++ {
res += 1.0
res += 2.5
res += 6.25
res += 15.625
}
On Friday, 4 August 2017 11:35:43 UTC+3, Egon wrote:
>
> Use the Assembly Luke.
>
> https://godbolt.org/g/nGFMbf
>
> It looks like clang manages to compute a table of powers of x. Which, is
> very very impressive.
>
> Which is roughly https://play.golang.org/p/CZkiJKfe7s -- except clang,
> also does inner loop unrolling.
>
> + Egon
>
> On Friday, 4 August 2017 11:16:10 UTC+3, Dorival Pedroso wrote:
>>
>> wait, what?!
>>
>> the same code:
>>
>> #include "stdio.h"
>> #include "math.h"
>> int main() {
>> double res = 0.0;
>> double x = 2.5;
>> int Nmax = 10000000;
>> for (int N=0; N<Nmax; N++) {
>> for (int i=0; i<20; i++) {
>> res += pow(x, i);
>> }
>> }
>> printf("res = %g\n", res);
>> }
>>
>> compiled with Clang is much faster?
>>
>> yes, I get:
>>
>> gcc -O2 c-code-sum.c -o c-code-sum_GCC -lm
>> time ./c-code-sum_GCC
>> res = 6.0633e+14
>>
>> real 0m8.210s
>> user 0m8.208s
>> sys 0m0.000s
>>
>> and:
>>
>> clang -O2 c-code-sum.c -o c-code-sum_CLANG -lm
>> time ./c-code-sum_CLANG
>> res = 6.0633e+14
>>
>> real 0m0.157s
>> user 0m0.156s
>> sys 0m0.004s
>>
>> So the question might be: what is CLANG doing? (and how to do this with
>> Go?...)
>>
>>
>> On Friday, August 4, 2017 at 6:04:01 PM UTC+10, Sebastien Binet wrote:
>>>
>>>
>>>
>>> On Fri, Aug 4, 2017 at 9:51 AM, Henrik Johansson <dahan...@gmail.com>
>>> wrote:
>>>
>>>> Actually I get the same as the original program on my mac.
>>>>
>>>> time ./ccode
>>>> sum=606329794183272.375000
>>>> ./ccode 0.17s user 0.00s system 98% cpu 0.170 total
>>>>
>>>> The Go version -O2 -Wall
>>>> time ./pow
>>>> sum=6.063297941832724e+14./pow 5.47s user 0.01s system 99% cpu 5.490
>>>> total
>>>>
>>>
>>> interesting :)
>>>
>>> I have:
>>>
>>> $> gcc --version
>>> gcc (GCC) 7.1.1 20170630
>>> Copyright (C) 2017 Free Software Foundation, Inc.
>>> This is free software; see the source for copying conditions. There is
>>> NO
>>> warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR
>>> PURPOSE.
>>>
>>> also interestingly, with clang:
>>> $> clang --version
>>> clang version 4.0.1 (tags/RELEASE_401/final)
>>> Target: x86_64-unknown-linux-gnu
>>> Thread model: posix
>>> InstalledDir: /usr/bin
>>>
>>> $> time ./c-code
>>>
>>> real 0m0.001s
>>> user 0m0.001s
>>> sys 0m0.000s
>>>
>>> $> time ./c-code-sum
>>> sum=606329794183272.375000
>>>
>>> real 0m0.228s
>>> user 0m0.227s
>>> sys 0m0.000s
>>>
>>> I haven't looked at the assembly to see what's really going on for the
>>> GCC version...
>>>
>>> -s
>>>
>>>
--
You received this message because you are subscribed to the Google Groups
"golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to golang-nuts+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
