[vector] binary operations with a scalar

[Paul Sandoz]

Hi Rahul,

Thanks for checking.

BTW i held off adding unit tests for the scalar cases. I was gonna wait for future updates from Jp w.r.t data providers.

Paul. 

> On Feb 23, 2018, at 12:00 PM, Kandu, Rahul <rahul.kandu at intel.com> wrote:
> 
> Hi Paul, John
> 
> I just checked that performance-wise there is no major concern with the binary ops accepting scalar values http://hg.openjdk.java.net/panama/dev/rev/3c6eb79d5312
> I took simple binary ops- Add, Mul and Sub for floats. Performance is same (ops/sec)- in both cases a) with broadcasting a scalar (outside the loop) vs. b) calling the scalar version of the method. Consider these simple loops for example. 
> 
>  static void AddTestWithScalar(float[] data1f, float data, float[] data3f) {
>        for (int i = 0; i < data1f.length; i += SPECIES.length()) {
>            FloatVector<Shapes.S256Bit> av = SPECIES.fromArray(data1f, i);
>           // FloatVector<Shapes.S256Bit> bv = SPECIES.fromArray(data2f, i);
>            av.add(data).intoArray(data3f, i);
>        }
>    }
> 
>    static void AddTestWithVectorHoisted(float[] data1f, float data, float[] data3f) {
>     FloatVector<Shapes.S256Bit> bv = SPECIES.broadcast(data);
>        for (int i = 0; i < data1f.length; i += SPECIES.length()) {
>            FloatVector<Shapes.S256Bit> av = SPECIES.fromArray(data1f, i);
>            av.add(bv).intoArray(data3f, i);
>        }
>    }
> 
> Rahul
> 
> -----Original Message-----
> From: panama-dev [mailto:panama-dev-bounces at openjdk.java.net] On Behalf Of John Rose
> Sent: Tuesday, February 20, 2018 1:16 PM
> To: Paul Sandoz <paul.sandoz at oracle.com>
> Cc: panama-dev at openjdk.java.net
> Subject: Re: [vector] binary operations with a scalar
> 
>> On Feb 20, 2018, at 9:19 AM, Paul Sandoz <paul.sandoz at oracle.com> wrote:
>> 
>> I like this but suggest we keep the code marginally simpler and revisit if we have performance concerns. If we don’t make intrinsic the broadcast+op combination i think something like this is probably ok for now:
>> 
>> @Override
>> @ForceInline
>> public Int256Vector add(int o) {
>>     return add(Int256Species.broadcastImpl(o));
>> }
> 
> That seems like the simplest way, and good to start with, maybe even end with. We could do decisively better with templates but we dint have them yet.