PROCESSOR-SDK-AM335X: Neon and VFPv3

Peter Mark77

I am trying to use some inline assembly but it won’t recognize any of the floating point instructions,
what are the correct flag to use both neon and vfp ?

SDK 5.1
QT Creator 4.6.2
arm-linux-gnueabihf-g++ (Linaro GCC 7.2-2017.11) 7.2.1 20171011

I have some code with neon intrinsics that seems to compile fine,
but this won’t ?

asm volatile (
"ldr %[db], s8 \n\t"
"ldr %[scale], s9 \n\t"
"fmov s11, #96.0 \n\t"
"fadd s8, s8, s9 \n\t"

{standard input}: Assembler messages:
{standard input}:2847: Error: bad instruction `fmov s11,#96.0'
{standard input}:2848: Error: bad instruction `fadd s8,s8,s9'

arm-linux-gnueabihf-g++ -c -include Meter1 -pipe -march=armv7-a -marm -mfpu=neon -mfloat-abi=hard --sysroot=/home/parallels/sitara/ti_sdk51/linux-devkit/sysroots/armv7ahf-neon-linux-gnueabi

over 5 years ago

0 RonB over 5 years ago

TI__Mastermind 30506 points

Peter,

These are generic ARM/GCC questions that TI does not support directly. We rely on ARM and GCC documentation to answer these as we pull them directly in and make no changes. Please prefer to the appropriate ARM and GCC docs for your questions.

Thank you.

0 Peter Mark77 over 5 years ago in reply to RonB

Prodigy 90 points

Ron,
thanks for your reply,
I did look at the documentation and tried different build settings but none are working,
so I was wondering if the compiler supplied with the SDK don't support it?

Thanks

0 RonB over 5 years ago in reply to Peter Mark77

TI__Mastermind 30506 points

Peter,

I'm sorry, I'm not a toolchain expert, but I would expect both neon and floating point to be supported (the hf stands for hard float). Whether they can both be used together and how to do it is the question I can't answer, sorry. We use the Linaro GCC and they have a lot of useful information at https://www.linaro.org/downloads/ and maybe that would be helpful to you.

Thanks.

Processors

Processors forum

PROCESSOR-SDK-AM335X: Neon and VFPv3