vecRevA used uninitialized in Levinson-Durbin MVE implementation

[Symanyshyn-IFX]

The Helium/MVE-optimized Levinson-Durbin functions reads from an uninitialized vector variable vecRevA on its first vsetq_lane_u64() call. The uninitialized lane value is discarded (overwritten on the next line), so there is no functional impact, but it is technically undefined behavior and triggers a warning on IAR.

Affected files:

CMSIS-DSP/Source/FilteringFunctions/arm_levinson_durbin_f32.c

Line 169 in e366b07

vecRevA = (f32x4_t) vsetq_lane_u64(tmpb,(uint64x2_t)vecRevA,0);

CMSIS-DSP/Source/FilteringFunctions/arm_levinson_durbin_f16.c

Line 166 in e366b07

vecRevA = (f16x8_t) vsetq_lane_u64(tmpb,(uint64x2_t)vecRevA,0);

CMSIS-DSP/Source/FilteringFunctions/arm_levinson_durbin_q31.c

Line 238 in e366b07

vecRevA = (q31x4_t) vsetq_lane_u64(tmpb,(uint64x2_t)vecRevA,0);

All three files have the same code pattern and the same bug.

Warning observed:

"arm_levinson_durbin_f32.c",169  Warning[Go029]: Variable "vecRevA" is used
    uninitialized at this location in the sourcecode.

"arm_levinson_durbin_f16.c",166  Warning[Go029]: Variable "vecRevA" is used
    uninitialized at this location in the sourcecode.

"arm_levinson_durbin_q31.c",238  Warning[Go029]: Variable "vecRevA" is used
    uninitialized at this location in the sourcecode.

Compiler: IAR C/C++ Compiler for Arm V9.70.1
Target: Cortex-M55
Build flags: --cpu Cortex-M55 -Ohz --fpu FPv5_D16 --aapcs vfp --endian=little -e --enable_restrict --no_wrap_diagnostics --no_dwarf4

[christophe0606]

@Symanyshyn-IFX It looks like it has already been corrected very recently : issue #298
The issue was about f16 but the commit also corrected f32 and q31.