[ Thread Index | Date Index | More lists.tuxfamily.org/eigen Archives ]
Advanced SIMD (aka ARMv8 NEON) support was added in commit 2164087:
I didn't know about Linaro's list to be honest, I just extended the
NEON port to 64-bit, as I had done the original NEON port a few years
ago . The 98% rate is actually acceptable, and I would consider the
port stable enough, but I don't really have a proper access to a ARMv8
system to run nightly tests and fix remaining bugs yet.
And you are right, since ARMv7 and ARMv8 use the same intrinsics for
NEON/Advanced SIMD, it was only a question of extending it to use
the new intrinsics for the 64-bit datatypes ([u]int64_t, double).
Btw, is there a public URL with the Linaro's list of software to be
ported to AArch64?
On Sat, 29 Nov 2014 18:34:02 -0500
Edwin Lum <pyourk@xxxxxxxxx> wrote:
> Hi all!
> I am participating in a project in which various open sourced software
> packages are being ported on to 64-bit ARM (Aarch 64). On Linaro's
> list of software packages that are to be ported, Eigen3 is one of
> them. I was able to run make check on eigen-3.2.2 using
> 3.17.1-302.fc21.aarch64 (Fedora 21). I do want to note that the CXX
> flag had to be set to g++ specifically, and that setting it to gcc
> does not work.
> I am including the output of make check to this email. The tests show
> a pass rate of 98%. Upon taking a look at dynalloc (the first test
> that failed), it seems that it requires valgrind which is not
> available on arm64 yet. The other tests fail usually with an abort. I
> get similar results on a x86 machine and get around 98% pass rate for
> the tests so I am going to assume that this is acceptable.
> From my understanding nothing special needs to be done on aarch64
> because of the nature of Eigen3, users take the needed header files
> and then compile them into the project they are working on. Is it
> okay to mark Eigen3 as functional on Aarch64 then?
|Mail converted by MHonArc 2.6.19+||http://listengine.tuxfamily.org/|