Claude review: better LUT segmentation for EOTFs

[Claude Code Review Bot <claude-review@example.com>]

Overall Series Review

Subject: better LUT segmentation for EOTFs
Author: Melissa Wen <mwen@igalia.com>
Patches: 6
Reviewed: 2026-05-07T13:11:56.356959

---

This is a well-motivated 5-patch series addressing gradient banding in HDR content (specifically PQ EOTFs) when AMD display hardware lacks fixed-function curves for certain color blocks (shaper, blend, regamma). The core insight is sound: the existing code used the same LUT segmentation helper (`cm3_helper_translate_curve_to_hw_format`) for both EOTFs (encoded→linear, degamma-like) and inverse EOTFs (linear→encoded, regamma-like), despite these having fundamentally different input domains and curvature distributions.

**Structure**: Patches 1-3 form the main fix — enabling GAMCOR for subsampled formats, reintroducing the degamma-specific translation helper, and wiring it up. Patches 4-5 are a precision refinement — adding interpolation and a halving segment distribution for PQ/sRGB. The author honestly notes patches 4-5 didn't produce visible improvement, which is refreshing transparency.

**Correctness**: The math checks out. Array bounds are respected (`rgb_resulted` has 259 entries, max `hw_points` is 256; `TRANSFER_FUNC_POINTS` is 1025, max tf_pts index accessed is 401). The `interp_tf_pts` interpolation is clean and correctly degenerates to a direct lookup when `seg_distr[k] <= 4`. The NULL-ramp path through `mod_color_calculate_degamma_params` is safe — all ramp dereferences are guarded by `ramp &&` checks.

**Scope**: Changes are tightly scoped to DCN3+ hardware (dcn32_hwseq.c), touching only 4 files. The series is reviewable and bisectable.

**Minor concerns**: One whitespace issue in patch 1, and the revert in patch 2 could benefit from a brief note about what was changed from the original removed code (if anything). Overall the series looks ready with minor fixes.

---

---
Generated by Claude Code Patch Reviewer