Fix ABF NODE_MEMREQ instance header underallocation

[goyalpalak18]

Description

I tracked down a nasty bug causing HardFaults on our ARM and RISC-V targets. We had a memory calculation mismatch in ee_abf_f32.c that was silently eating our CMSIS-DSP safety padding. It was invisible on standard x86 builds, but broke on target hardware with strict memory protection (MPU/PMP) enabled.

Root Cause

In the NODE_MEMREQ calculation, we were only budgeting for two pointers. However, beamformer_f32_reset() actually uses all four pointers defined in the abf_f32_instance_t struct.

Because we shortchanged the allocation by 8 to 16 bytes, the instance header spilled over and consumed the 12-byte CMSIS-DSP safety padding at the end of the buffer. When vectorized operations (like Helium or NEON) did their normal read-past-end tail processing, they hit unmapped memory and triggered immediate HardFaults.

The Fix

I replaced the hardcoded pointer math with sizeof(abf_f32_instance_t). This correctly accounts for the entire struct, is self-documenting, and prevents the header from overwriting the safety padding.

Impact

• Stability: Eliminates the HardFaults/crashes on our high-performance embedded targets.
• Memory: Fully restores the required CMSIS-DSP safety padding (ensuring 16 bytes of padding on both 32-bit and 64-bit systems). Total heap allocation just goes up by a harmless 8-16 bytes.
• Algorithm: Completely untouched. Data flow and output remain exactly the same.

[goyalpalak18]

@llefaucheur @joseph-yiu I put together a fix for the memory allocation issue in ee_abf_f32.c. Ready for review when you have a moment

[llefaucheur]

Its good for me. I did not checked but I see no reason the Audiomark score to being changed with this patch.

[goyalpalak18]

Thanks @llefaucheur . I can confirm the scores will stay exactly the same. Let me know if you need anything else before merging!

[joseph-yiu]

Hi @goyalpalak18 , @llefaucheur ,
I am away right now (holiday) and can help reviewing and merging this into the dev branch when I am back next week.
regards,
Joseph

[joseph-yiu]

Merged into dev branch (dev_2026q1). Thanks for the patch.
regards,
Joseph

[joseph-yiu]

Closing this pull request as the AudioMark v1.0.4 release is completed.