On Tue, Jul 9, 2013 at 4:54 PM, Alexander Chemeris
<[email protected]> wrote:
>> From: Thomas Tsou <[email protected]>
>> Attached are the standalone unit test cases for SSE 4.2. As previously
>> mentioned, Atom needs SSE3 only. I'll add the ifdefs for those
>> shortly. I don't know if there's an appropriate repository for these
>> right now - linking libosmocore from the transceiver for comparison
>> purposes only seems silly. I just generated a temporary tarball for
>> the time being.
>
> Ok, waiting for the updated version to test it on my Core 2 Duo and Atoms.
Pushed for the time being to:
https://github.com/ttsou/sse-tests.git
To enable SSE4:
./configure --with-sse4
Vector 16-bit integer to floating point conversion is only enabled
with SSE4. In general, the SSE type conversion benefits are marginal
because the native conversion instructions operate on 32-bit widths,
which requires a separate sign extension. If there is no dedicated
instruction for sign extension (SSE3) then the benefits (if any) are
reduced even more.
Thomas
