linuxaudio.org

--_bf59ea93-de70-435b-ae29-43d7d16beb3c_

Content-Type: text/plain; charset="iso-8859-1"

Content-Transfer-Encoding: quoted-printable
> No=2C when building with Android NDK using the armeabi-v7a ABI=2C hard fl=

oats are

quoted text
> used. This works on ARMv7=2C which means a lot of devices=2C and certainl=

y the

quoted text
> majority.

But then every other system gets the rough 1000 factor performance hit that=

 started

this thread since all other CPU have to do a system call to execute the flo=

at operation.

Designing for a single handset family is not that scalable.
> > The ARM

quoted text
> > softfloat overhead is not that great and the coding required to get

> > access to the GPUs is suitable that developers will implement them

> > for optimisations.

>=20

> Soft floats are insanely slow on Android in my experience.

A factor of 10 or 20 over hard floats? Sound about right? The factor 1000 i=

s for code

that uses FPU instructions on systems without the FPU support. That I would=

 call

insanely slow but you can choose which way you go here.
> You got an FPU on most devices=2C you can compile for hard floats=2C acti=

vate ARM

quoted text
> NEON if you wish too=2C and even check for cpu features at runtime. I don=

't see

quoted text
> why you want to use the GPU for optimization.

Man=2C the recent chipsets have 6 of these pipelines - designed for graphic=

s processing

but sitting there waiting for audio DSP.
Regards=2C nick.

 		 	   		  =
--_bf59ea93-de70-435b-ae29-43d7d16beb3c_

Content-Type: text/html; charset="iso-8859-1"

Content-Transfer-Encoding: quoted-printable
&gt=3B No=2C when building with Android NDK using the armeabi-v7a ABI=2C ha=

rd floats are&gt=3B used. This works on ARMv7=2C which means a lot=

 of devices=2C and certainly the&gt=3B majority.But then every =

other system gets the rough 1000 factor performance hit that startedthi=

s thread since all other CPU have to do a system call to execute the float =

operation.Designing for a single handset family is not that scalable.&gt=3B &gt=3B The ARM&gt=3B &gt=3B softfloat overhead is not that=

 great and the coding required to get&gt=3B &gt=3B access to the GPUs i=

s suitable that developers will implement them&gt=3B &gt=3B for optimis=

ations.&gt=3B &gt=3B Soft floats are insanely slow on Android in my=

 experience.A factor of 10 or 20 over hard floats? Sound about righ=

t? The factor 1000 is for codethat uses FPU instructions on systems wit=

hout the FPU support. That I would callinsanely slow but you can choose=

 which way you go here.&gt=3B You got an FPU on most devices=2C you=

 can compile for hard floats=2C activate ARM&gt=3B NEON if you wish too=

=2C and even check for cpu features at runtime. I don't see&gt=3B why y=

ou want to use the GPU for optimization.Man=2C the recent chipsets =

have 6 of these pipelines - designed for graphics processingbut sitting=

 there waiting for audio DSP.Regards=2C nick. 		 	   		  =
=
--_bf59ea93-de70-435b-ae29-43d7d16beb3c_--