built-in string hash ?

[BCS]

Hello bearophile,

> Pelle:
> 
>> I doubt that gives any performance gains. typeid(a).getHash should be
>> a constant expression anyway, and I don't see any gains in my tiny
>> benchmark test.
>> 
> My code shows a limited time difference:
> 
> import std.stdio: writeln;
> import std.date: getUTCtime, ticksPerSecond;
> void main() {
> enum int NLOOP = 30_000_000;
> string s = "and I'd like to add a toHash() member, but I can't
> find the built-in string hash function.";
> {
> auto t0 = getUTCtime();
> auto stringHash =&(typeid(s).getHash);
> hash_t tot;
> for (int i; i < NLOOP; i++)
> tot += stringHash(&s);
> auto t1 = getUTCtime();
> writeln(tot, " ", (t1 - t0) / cast(double)ticksPerSecond);
> }
> {
> auto t0 = getUTCtime();
> hash_t tot;
> for (int i; i < NLOOP; i++)
> tot += typeid(s).getHash(&s);
> auto t1 = getUTCtime();
> writeln(tot, " ", (t1 - t0) / cast(double)ticksPerSecond);
> }
> }
> And the asm shows the first loop to contain one less instruction (dmd
> 2.048, -O -release -inline), so the difference is small:
> 
> L42:    lea ECX,01Ch[ESP]
> mov EAX,02Ch[ESP]
> push    ECX
> call    EDI
> add ESI,EAX
> inc EBX
> cmp EBX,01C9C380h
> jb  L42
> LA7:    lea EDX,01Ch[ESP]
> mov ECX,_D12TypeInfo_Aya6__initZ

DMD seems to do a bad job of dead assignment removal at the ASM level. I've 
seen this befor where half the asm in large blocks was dead.

> mov EAX,offset FLAT:_D12TypeInfo_Aya6__initZ
> push    EDX
> call    dword ptr 018h[ECX]
> add EDI,EAX
> inc EBX
> cmp EBX,01C9C380h
> jb  LA7
> Bye,
> bearophile
-- 
... <IXOYE><