CODEWITHSUNDEEP
assemblyarmneoncortex-mkeil
Et8
Et8

Reputation: 1

VCVT instruction on ARM M4-F to convert from floating point to fixed point not recognized

I am trying to use the assembly instruction "vcvt.s32.f32 q8, q8, #0xf" in an ARM Cortex M4-F using the Arm Keil IDE and the compiler versions v5 and v6.

Unfortunately, this instruction is not recognized by the compiler. When I write C code, the compiler uses other slower assembly instructions in place of this one; If I directtly write the assembly instruction, it is not accepted by the compiler leading to a compiling error.

Does anyone know anything about it?

This instruction looks present in the ARM M4-F ISA, but currently I am not able to make it works!

Upvotes: 0

Views: 1164

Answers (1)

Peter Cordes
Peter Cordes

Reputation: 364210

It assembles with the GNU assembler using arm-none-eabi-as -mfpu=neon -mcpu=cortex-m4 -mthumb foo.s in case that helps.

But not with -mcpu=cortex-m4 -mfpu=vfpv4.

Wikipedia says that the optional FPU on M4 is VFPv4-SP (single precision), which I think also implies that it's not full NEON. Tom V also confirms in comments that Cortex-M4-F doesn't support this NEON SIMD instruction.

Keil is correct -
Cortex-M4-F doesn't support vcvt.s32.f32 for q or d registers, only for s registers.

The following does assemble ok with
arm-none-eabi-as -mfpu=vfpv4 -mcpu=cortex-m4 -mthumb:

   vcvt.s32.f32 s8, s8, #0xf
   vcvt.s32.f32 s9, s9, #0xf
   vcvt.s32.f32 s10, s10, #0xf
   vcvt.s32.f32 s11, s11, #0xf

Upvotes: 1

Related Questions