Replies to post #130421 on Intel Corp (INTC)

[ibc]

Just like x86 has FTZ mode for SSE2 floating point.

[Dmcq]

For VFP9 and VFP10, ARM has "full IEEE754 compliance with ARM support software, or near IEEE754 compliance with hardware only". I suspect the support software needed for full IEEE754 compliance causes a performance impact when underflow occurs or denormals are loaded from memory.

My understanding is they support all that fully in hardware.

The problem they have had is with the SIMD which only does flush to zero, it was really built as a support for graphics and suchlike and not for user maths processing. This has been fixed in the 64 bit architecture. They have become quite hot on standards so even their later graphics units do this okay - they probably share the hardware design is my guess considering their resources ;-)

I think the near full refers to operations like floating point remainder divide needing to be done in software.