Intel AMX and feature detection

[Paul Sandoz]

Is it possible to share further details on a new email thread as to what might be causing that?

In this context I think John is more referring to the case of no compiler support.

There should support on SSE2/3, same for NEON, and SVE when 128 bit (I think for NVIDIA Grace CPUs), but we may not have all paths covered optimally on all the platforms.

Paul.

> On Jun 27, 2024, at 2:44 AM, Robert Muir <rcmuir at gmail.com> wrote:
> 
> On Wed, Jun 26, 2024 at 7:03 PM John Rose <john.r.rose at oracle.com> wrote:
>> 
>> Actually we have VectorShape.S_64_BIT which, if it is
>> the preferred shape, is really telling you the
>> “vector” processing is inside the CPU not the VPU.
>> That’s a good-enough hint to avoid the Vector API,
>> right?
> 
> That's not what I see. On an x86 machine without AVX, both FloatVector
> and IntVector say 128-bit, but if we use IntVector it results in 30x
> slower lucene :(