Please rid me of this goto

[Seb via Digitalmars-d]

On Thursday, 23 June 2016 at 21:03:08 UTC, Andrei Alexandrescu 
wrote:
> On 06/23/2016 02:37 PM, Seb wrote:
>> On Thursday, 23 June 2016 at 18:05:07 UTC, Andrei Alexandrescu 
>> wrote:
>>> On 06/23/2016 01:34 PM, H. S. Teoh via Digitalmars-d wrote:
>>>> I don't understand why that goto is necessary.
>>>
>>> Eh, thank you all who set me straight! I've been in my head 
>>> for too
>>> long. So where is the current implementation of "^^"? If it's 
>>> not as
>>> fast as this, we should replace it. -- Andrei
>>
>> ^^ seems to be lowered here
>> https://github.com/dlang/dmd/blob/9903aba3b1d39bf499a54edbc81c7d9f08711f3e/src/constfold.d#L506
>>
>>
>> and std.math.pow is defined here:
>> https://github.com/dlang/phobos/blob/master/std/math.d#L6028
>
> I see, thanks. Both seem to be ripe for the optimized version 
> (they do more testing than necessary).
>
>> However you should test how it performs against the LLVM 
>> intrinsics
>> available in LDC, e.g.:
>>
>> llvm.intrinsincs.llvm_pow and llvm_powi (i = integer).
>
> Cool! Is that a CPU operation?

Depends on the target
http://llvm.org/docs/LangRef.html#llvm-pow-intrinsic


You should definitely benchmark between compilers. I gave it a 
quick for floating point and integer power (see [1, 2]). Please 
take the numbers with _great_ care, but I hope they show the 
point about the importance of benchmarking between compilers ;-)

Floating-point
--------------

dmd -inline -release -O -boundscheck=off test_pow.d 
-ofbin/dmd/test_pow
ldc -release -O3 -boundscheck=off test_pow.d -ofbin/ldc/test_pow
gdc -finline-functions -frelease -O3 test_pow.d -o 
bin/gdc/test_pow

>dmd
math.pow 3 secs, 463 ms, 379 μs, and 6 hnsecs
^^       7 secs, 295 ms, 984 μs, and 3 hnsecs

>ldc
llvm_pow 125 ms, 673 μs, and 2 hnsecs
^^        6 secs, 380 ms, 104 μs, and 2 hnsecs

>gdc
math.pow 0 secs 125 ms
^^       2 secs 2161 ms

Integer power
-------------

dmd -inline -release -O -boundscheck=off test_powi.d 
-ofbin/dmd/test_powi
ldc -release -O3 -boundscheck=off test_powi.d -ofbin/ldc/test_powi
gdc -finline-functions -frelease -O3 test_powi.d -o 
bin/gdc/test_powi

>dmd
math.pow 978 ms, 961 μs, and 8 hnsecs
^^       1 sec, 15 ms, 177 μs, and 6 hnsecs

>ldc
math.pow 556 ms, 366 μs, and 8 hnsecs
^^       507 ms, 3 μs, and 1 hnsec

>gdc
math.pow 0 secs 125 ms
^^       0 secs 42 ms


[1] https://github.com/wilzbach/perf-d/blob/master/test_pow.d
[2] https://github.com/wilzbach/perf-d/blob/master/test_powi.d