asm clean up and stack size reduction

[andrewstanfordjason]

This PR cleans up the lib_xcore_math filter code for vx4b and reduces the stack usage to its minimum. It also fixes a bug in the filter tests.

[Copilot]

Pull request overview

This PR cleans up and optimises the VX4B filter assembly implementations in lib_xcore_math, primarily by reducing stack usage, and adjusts tests to match the VX4B numerical behaviour and fix a shift-handling bug in the FIR S32 tests.

Changes:

• Reduce NSTACKWORDS and adjust stack layouts for VX4B filter assembly routines (filter_fir_s32, filter_fir_s16, biquad variants, and support routines) to minimise stack usage and remove old tool-annotation clutter.
• Fix the S32 FIR test’s expected-value computation for shift == 0 and relax several S16 FIR tests on __VX4B__ to allow small LSB differences.
• Clean up VX4B VPU usage in various filter assembly files (use of la for constants, updated mask and pointer handling, consistent .globl/metadata directives).

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/filter_tests/src/filter/test_filter_fir_s32.c	Adjusts expected-value logic to treat shift == 0 as a non-shift (avoiding undefined (1LL << -1)), fixing the S32 FIR test bug.
tests/filter_tests/src/filter/test_filter_fir_s16.c	For __VX4B__, replaces exact equality checks with small INT16_WITHIN tolerances to account for VX4B rounding/saturation differences.
lib_xcore_math/src/arch/vx4b/filter/vect_s32_convolve_valid.S	Reduces stack usage and removes translation warnings while keeping the convolution loop and tail handling logic intact.
lib_xcore_math/src/arch/vx4b/filter/push_sample_up_s16.S	Shrinks the stack frame, fixes prologue/epilogue register save/restore offsets, and updates comments and branch usage for VX4B.
lib_xcore_math/src/arch/vx4b/filter/push_sample_down_s16.S	Similar stack-frame reduction and cleanup for the “push sample down” helper, with corrected save/restore and VPU loop structure.
lib_xcore_math/src/arch/vx4b/filter/filter_fir_s32.S	Reduces stack usage and cleans up the VPU setup, tail processing, and accumulator-combination logic, but currently has an uninitialised s4 used as a shift amount in the left-shift path.
lib_xcore_math/src/arch/vx4b/filter/filter_fir_s16.S	Refactors stack layout and adds extra vector scratch regions for VX4B, but currently computes a stack-based pointer (s7) that underflows the frame and is then used by VPU load/store instructions.
lib_xcore_math/src/arch/vx4b/filter/filter_biquad_sat_s32.S	Tightens the stack frame and vector scratch allocation while preserving the biquad-saturated processing pipeline and state-update logic.
lib_xcore_math/src/arch/vx4b/filter/filter_biquad_s32.S	Minimises stack usage and simplifies the non-saturating biquad filter implementation, updating VPU constant loads and state propagation.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[andrewstanfordjason]

@copilot please review

[Copilot]

@andrewstanfordjason I've opened a new pull request, #210, to work on those changes. Once the pull request is ready, I'll request review from you.

[Copilot]

Pull request overview

Copilot reviewed 9 out of 9 changed files in this pull request and generated 1 comment.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.