STM32F4 compiler

[developer2]

Posted on April 22, 2014 at 22:23

Hi all,

does anybody have running gcc with simd for Cortex-M4 ? Kind regards,

[Tesla DeLorean]

Posted on April 24, 2014 at 13:45

does anybody have running gcc with simd for Cortex-M4 ?

Can't say I've tried, but the hardware FPU stuff seems to work. Perhaps there are some more GNU/GCC centric forums or lists?

[frankmeyer9]

Posted on April 24, 2014 at 13:56

does anybody have running gcc with simd for Cortex-M4 ?

I'm not aware that the M4 supports any SIMD instructions. The (optional) NEON unit on Cortex A devices does.

[chen]

Posted on April 24, 2014 at 16:41

Hi

The programmers reference manual says that it does. https://my.st.com/sites/xnet_docs/drc/drc.aspx?documenttype=Programming%20Manual&prodtype=SS&prodid=1577&count=2 It only works on some Mult instructions. Not sure how much performance gain it is going to get!?!

[waclawek.jan]

Posted on April 24, 2014 at 17:39

Have a look at core_cm4_simd.h in CMSIS directory of your toolchain.

[EDIT] Apparently gcc does use ''natively'' at least some of them, e.g. uint32_t aa(int32_t x, int32_t y) {   int64_t acc;   acc = x;   acc += (int64_t) x * y;   return (acc >> 32); } gcc 4.7.0 compiled into 0800019c <aa>:  800019c:    17c3          asrs    r3, r0, #31  800019e:    4602          mov    r2, r0  80001a0:    fbc1 2300     smlal    r2, r3, r1, r0  80001a4:    4618          mov    r0, r3  80001a6:    4770          bx    lr [EDIT2] Ah, smlal is not considered SIMD, sorry... 🙂 JW

[frankmeyer9]

Posted on April 24, 2014 at 20:31

Albeit even ARM talks about ''optimized SIMD ... instructions'', I view this as a rather overstretched definition of SIMD.

Just compare this with NEON or other MME/GPU unit, doing 32 ... 512 identical operations in parallel in a single instruction.

[developer2]

Posted on April 24, 2014 at 20:39

fm ... please could you give me the hint ... is there processor in QFP package ? 

[frankmeyer9]

Posted on April 24, 2014 at 21:11

Don't know, I'm no hardware designer. I would have to use a search engine, too.

NXP had produced LPC8xx Cortex-M0 MCUs in DIL package lately, so they might have QFP, too. Recently I am dealing with the x86 compatible embedded world, where everything is BGA...