Fix CY0 buffer overflow in beamformer working struct

[goyalpalak18]

Summary

Identified and fixed a buffer overflow in the CY0 scratch buffer.

The buffer was declared using the half-spectrum constant (NFFTD2), but the synthesis logic treats it as full-spectrum (NFFT). This resulted in writing 126 floats (504 bytes) past the end of the array during the IFFT and spectrum extension stages.

On the current ARM build, this overflow was overwriting the adjacent XY struct member. It did not trigger a crash solely due to the specific memory layout and the fact that XY was effectively unused at that point in the frame. This fix updates CY0 to the correct full-spectrum size to prevent the out-of-bounds writes.

The Bug

The CY0 buffer was declared with NFFTD2 (half-spectrum size), likely because it's used for some half-spectrum math early in the function.

However, later in the synthesis stage (specifically the complex IFFT), we treat CY0 as a full-spectrum buffer. We end up writing about 126 floats past the end of the array.

The Fix

I just bumped the size of CY0 to match the other full-spectrum buffers (mic, BM, etc).

// src/ee_abf_f32.h
- ee_f32_t CY0[NFFTD2 * COMPLEX + 2]; // Too small (130 floats)
+ ee_f32_t CY0[NFFT * COMPLEX + 2];   // Correct (258 floats)

[goyalpalak18]

Hi @joseph-yiu @petertorelli,

This PR fixes a buffer overflow in the beamformer where CY0 was declared using NFFTD2 (half size) but used as NFFT (full size) in the synthesis loop. This was causing a 504-byte write past the end of the array.

I verified the fix with AddressSanitizer and confirmed that benchmark scores match the baseline.

I’d appreciate a review when you get a chance. Happy to make any changes if needed.

Thanks!

[joseph-yiu]

Thanks the for pull request. We will review.
regards,
Joseph

[joseph-yiu]

Bug confirmed and the code changes have been review. I merged the PR into dev_20261q1 branch.
regards,
Joseph