Stencil blendingmode and some speed optimisations #4889
Conversation
- better performance at the cost of about 700 bytes of flash
- Stencil overlay adds an FX on top but leaves the backgroundcolor transparent, this is useful for classic text overlays
WalkthroughUnifies per-pixel blending into a new segblend routine, adds stencil (bm=32) support, introduces push/mirror/reverse/transpose coordinate handling, updates the UI to send explicit blend values, and simplifies JSON deserialization for segment blendMode; header docs updated to mention "stencil". Changes
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes Possibly related PRs
Suggested reviewers
✨ Finishing Touches
🧪 Generate unit tests
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. 🪧 TipsChatThere are 3 ways to chat with CodeRabbit:
SupportNeed help? Create a ticket on our support page for assistance with any issues or questions. CodeRabbit Commands (Invoked using PR/Issue comments)Type Other keywords and placeholders
CodeRabbit Configuration File (
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (3)
wled00/FX.h (1)
459-459: Document exact bm codes, incl. Stencil (32) and Average (5).Make the mapping explicit to avoid confusion and to reflect the UI: 0=top, 1=bottom, 2=add, 3=subtract, 4=difference, 5=average, 6=multiply, 7=divide, 8=lighten, 9=darken, 10=screen, 11=overlay, 12=hardlight, 13=softlight, 14=dodge, 15=burn, 32=stencil (uses colors[1] as background mask).
- uint8_t blendMode; // segment blending modes: top, bottom, add, subtract, difference, multiply, divide, lighten, darken, screen, overlay, hardlight, softlight, dodge, burn, stencil + uint8_t blendMode; // segment blending modes (bm): 0=top,1=bottom,2=add,3=subtract,4=difference,5=average,6=multiply,7=divide,8=lighten,9=darken,10=screen,11=overlay,12=hardlight,13=softlight,14=dodge,15=burn,32=stencil (bg=colors[1])wled00/data/index.js (1)
2362-2365: Send bm as a number, not a string.Parse to int before sending to avoid accidental string types and keep consistency with other setters.
- var value = gId(`seg${s}bm`).value; - var obj = {"seg": {"id": s, "bm": value}}; + const value = parseInt(gId(`seg${s}bm`).value, 10); + const obj = {"seg": {"id": s, "bm": value}};wled00/json.cpp (1)
305-305: Validate bm to allowed set (0–15, 32).Removing the clamp enables arbitrary values. segblend defaults to “Top”, but guarding here prevents unexpected UI/API inputs from drifting.
- getVal(elem["bm"], seg.blendMode); + if (getVal(elem["bm"], seg.blendMode)) { + if (seg.blendMode != 32 && seg.blendMode > 15) seg.blendMode = 0; // fallback to Top + }
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
💡 Knowledge Base configuration:
- MCP integration is disabled by default for public repositories
- Jira integration is disabled by default for public repositories
- Linear integration is disabled by default for public repositories
You can enable these sources in your CodeRabbit configuration.
📒 Files selected for processing (4)
wled00/FX.h(1 hunks)wled00/FX_fcn.cpp(6 hunks)wled00/data/index.js(2 hunks)wled00/json.cpp(1 hunks)
🧰 Additional context used
📓 Path-based instructions (2)
wled00/data/**/*.{htm,html,css,js}
📄 CodeRabbit inference engine (.github/copilot-instructions.md)
Use tabs for indentation in web files (.htm/.html/.css/.js) under wled00/data
Files:
wled00/data/index.js
wled00/{*.cpp,!(html_*)*.h}
📄 CodeRabbit inference engine (.github/copilot-instructions.md)
Use spaces (2 per level) for C++ source and header files
Files:
wled00/FX.hwled00/json.cppwled00/FX_fcn.cpp
🧠 Learnings (2)
📚 Learning: 2025-08-26T11:51:21.817Z
Learnt from: DedeHai
PR: wled/WLED#4798
File: wled00/FX.cpp:7531-7533
Timestamp: 2025-08-26T11:51:21.817Z
Learning: In WLED PR #4798, DedeHai confirmed that certain gamma-related calls in FX.cpp/FX_fcn.cpp/particle systems are intentional for effect-level shaping (e.g., brightness curves, TV sim, Pride 2015 pre-mix), distinct from final output gamma. Do not flag or remove these in future reviews; add comments when feasible to clarify intent.
Applied to files:
wled00/FX_fcn.cpp
📚 Learning: 2025-06-07T15:58:42.579Z
Learnt from: willmmiles
PR: wled/WLED#4712
File: wled00/FX_fcn.cpp:992-993
Timestamp: 2025-06-07T15:58:42.579Z
Learning: In WLED segments, the backing pixel buffer is always allocated to the full physical segment size, but effect rendering only uses a subset when mirror or grouping are enabled. For 2D segments, this subset is vWidth() * vHeight(), and for 1D segments it's vLength(). Color processing operations like fade_out should iterate over this rendering subset, not the full allocated buffer.
Applied to files:
wled00/FX_fcn.cpp
🧬 Code graph analysis (2)
wled00/json.cpp (1)
wled00/util.cpp (2)
getVal(66-83)getVal(66-66)
wled00/FX_fcn.cpp (1)
wled00/colors.cpp (4)
color_add(28-61)color_add(28-28)color_blend(11-21)color_blend(11-11)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (18)
- GitHub Check: wled_build / Build Environments (usermods)
- GitHub Check: wled_build / Build Environments (esp32s3dev_8MB_opi)
- GitHub Check: wled_build / Build Environments (esp32s3_4M_qspi)
- GitHub Check: wled_build / Build Environments (esp32c3dev)
- GitHub Check: wled_build / Build Environments (esp32s3dev_16MB_opi)
- GitHub Check: wled_build / Build Environments (esp01_1m_full_compat)
- GitHub Check: wled_build / Build Environments (esp01_1m_full_160)
- GitHub Check: wled_build / Build Environments (nodemcuv2_160)
- GitHub Check: wled_build / Build Environments (esp32dev)
- GitHub Check: wled_build / Build Environments (esp32_eth)
- GitHub Check: wled_build / Build Environments (nodemcuv2_compat)
- GitHub Check: wled_build / Build Environments (esp32_wrover)
- GitHub Check: wled_build / Build Environments (esp01_1m_full)
- GitHub Check: wled_build / Build Environments (esp8266_2m_160)
- GitHub Check: wled_build / Build Environments (lolin_s2_mini)
- GitHub Check: wled_build / Build Environments (esp8266_2m_compat)
- GitHub Check: wled_build / Build Environments (nodemcuv2)
- GitHub Check: wled_build / Build Environments (esp8266_2m)
🔇 Additional comments (2)
wled00/data/index.js (1)
828-829: UI option for Stencil (32) looks good.Matches FX_fcn implementation and JSON handling.
wled00/FX_fcn.cpp (1)
1345-1572: Consolidation into segblend + push/mirror logic: solid.The restructure reduces branching and enables Stencil; coordinate transforms are gated and reused cleanly.
There was a problem hiding this comment.
Actionable comments posted: 0
♻️ Duplicate comments (1)
wled00/FX_fcn.cpp (1)
1324-1324: Subtract mode green-channel bug fixedThe green channel now subtracts bottom from top like the others. Thanks for addressing the prior issue.
🧹 Nitpick comments (3)
wled00/FX_fcn.cpp (3)
1317-1343: segblend() consolidation: clear and fast pathSingle switch-based routine reduces indirections and keeps hot code tight. Consider forcing placement/inlining for hot path:
- Add IRAM attribute and inline hint to avoid flash stalls in tight per-pixel loops.
-static uint32_t segblend(CRGBW tcol, CRGBW bcol, CRGBW bgcol, uint8_t blendMode) { +IRAM_ATTR static inline uint32_t segblend(CRGBW tcol, CRGBW bcol, CRGBW bgcol, uint8_t blendMode) {
1339-1340: Stencil semantics: exact color match vs. toleranceEquality on 32-bit color is correct for a pure stencil, but can be brittle if upstream effects introduce minor channel drift. If you see unexpected “holes,” consider (optional) channel-tolerance compare.
- case 32: return tcol.color32 == bgcol.color32 ? bcol.color32 : tcol.color32; // Stencil + case 32: { + // Optional tolerance (e.g. 0..2) to avoid miss due to rounding/dithering + constexpr uint8_t tol = 0; + bool isBG = + (abs(int(tC.r)-int(bgcol.r)) <= tol) && + (abs(int(tC.g)-int(bgcol.g)) <= tol) && + (abs(int(tC.b)-int(bgcol.b)) <= tol) && + (abs(int(tC.w)-int(bgcol.w)) <= tol); + return isBG ? bcol.color32 : tcol.color32; + }
1452-1463: Push offsets computed once; clean modulo handlingEfficient offset math and sign handling. Minor nit: make offset vars int to avoid implicit unsigned wrap on negatives before the modulo.
- unsigned offsetX = (blendingStyle == BLEND_STYLE_PUSH_UP || blendingStyle == BLEND_STYLE_PUSH_DOWN) ? 0 : progInv * nCols / 0xFFFFU; - unsigned offsetY = (blendingStyle == BLEND_STYLE_PUSH_LEFT || blendingStyle == BLEND_STYLE_PUSH_RIGHT) ? 0 : progInv * nRows / 0xFFFFU; + int offsetX = (blendingStyle == BLEND_STYLE_PUSH_UP || blendingStyle == BLEND_STYLE_PUSH_DOWN) ? 0 : int(progInv * nCols / 0xFFFFU); + int offsetY = (blendingStyle == BLEND_STYLE_PUSH_LEFT || blendingStyle == BLEND_STYLE_PUSH_RIGHT) ? 0 : int(progInv * nRows / 0xFFFFU);
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
💡 Knowledge Base configuration:
- MCP integration is disabled by default for public repositories
- Jira integration is disabled by default for public repositories
- Linear integration is disabled by default for public repositories
You can enable these sources in your CodeRabbit configuration.
📒 Files selected for processing (1)
wled00/FX_fcn.cpp(8 hunks)
🧰 Additional context used
📓 Path-based instructions (1)
wled00/{*.cpp,!(html_*)*.h}
📄 CodeRabbit inference engine (.github/copilot-instructions.md)
Use spaces (2 per level) for C++ source and header files
Files:
wled00/FX_fcn.cpp
🧠 Learnings (2)
📚 Learning: 2025-08-26T11:51:21.817Z
Learnt from: DedeHai
PR: wled/WLED#4798
File: wled00/FX.cpp:7531-7533
Timestamp: 2025-08-26T11:51:21.817Z
Learning: In WLED PR #4798, DedeHai confirmed that certain gamma-related calls in FX.cpp/FX_fcn.cpp/particle systems are intentional for effect-level shaping (e.g., brightness curves, TV sim, Pride 2015 pre-mix), distinct from final output gamma. Do not flag or remove these in future reviews; add comments when feasible to clarify intent.
Applied to files:
wled00/FX_fcn.cpp
📚 Learning: 2025-06-07T15:58:42.579Z
Learnt from: willmmiles
PR: wled/WLED#4712
File: wled00/FX_fcn.cpp:992-993
Timestamp: 2025-06-07T15:58:42.579Z
Learning: In WLED segments, the backing pixel buffer is always allocated to the full physical segment size, but effect rendering only uses a subset when mirror or grouping are enabled. For 2D segments, this subset is vWidth() * vHeight(), and for 1D segments it's vLength(). Color processing operations like fade_out should iterate over this rendering subset, not the full allocated buffer.
Applied to files:
wled00/FX_fcn.cpp
🧬 Code graph analysis (1)
wled00/FX_fcn.cpp (1)
wled00/colors.cpp (4)
color_add(28-61)color_add(28-28)color_blend(11-21)color_blend(11-11)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (5)
- GitHub Check: wled_build / Build Environments (esp32s3_4M_qspi)
- GitHub Check: wled_build / Build Environments (usermods)
- GitHub Check: wled_build / Build Environments (esp32dev)
- GitHub Check: wled_build / Build Environments (lolin_s2_mini)
- GitHub Check: wled_build / Build Environments (esp32s3dev_16MB_opi)
🔇 Additional comments (8)
wled00/FX_fcn.cpp (8)
371-374: Palette blend clamp looks goodThe explicit 255 cap avoids pathological loops; safe and clear.
1345-1351: Propagating bgColor for stencil-enabled blendingPlumbing secondary color as background for modes that need it is correct; keeps segblend signature minimal.
1425-1447: 2D mirroring writes: order and bounds look correct
- Single write + mirrored writes use segblend then opacity mix; consistent with 1D path.
- CCT mirror updates are guarded and aligned with pixel writes. LGTM.
1455-1456: applyReverse/applyMirror flags gate transformsGood micro-optimization to avoid unnecessary per-pixel conditionals when no reverse/transpose/mirror is active.
1474-1475: Push mapping before sampling old/new segmentsCorrect to apply push before sampling seg/segO so clipped/fade logic sees the shifted canvas consistently.
1492-1495: Reverse/transpose only when neededDeferred transform after sampling avoids double-mapping. Clean and readable.
1518-1531: 1D path parity with 2D: segblend + opacityMirroring, wrapping with offset, and segblend usage match 2D semantics. Consistency reduces edge-case drift. LGTM.
1613-1615: Gamma application guard aligns with realtime overrideSkipping gamma when ARLS requests it is correct. Note: per prior learnings, effect-level gamma shaping elsewhere is intentional—no changes needed here.
|
A few more comments: The use of CRGBW instead of uint32_t reduces flash size and increases speed, although how exactly eludes me: the two aliases tC / bC don't do anything but make code more compact to read and came from a version where I passed uint32_t and then cast that: if instead of aliases the arguments are used directly, codesize increases by 250bytes, no idea why having aliases makes the compiler happy. one more peculiarity: when I did initial tests in a different branch (heap_handling), using "O2" optimization flag on |
IIRC, often big changes in code size are caused by loop unrolling. |
wled00/FX_fcn.cpp
Outdated
| case 0: return tcol.color32; // Top | ||
| case 1: return bcol.color32; // Bottom | ||
| case 2: return color_add(tcol, bcol, false); // Add | ||
| case 3: c.r = _subtract(tC.r, bC.r); c.g = _subtract(tC.g, bC.g); c.b = _subtract(tC.b, bC.b); c.w = _subtract(tC.w, bC.w); break; // Subtract |
There was a problem hiding this comment.
A bit of a nitpick, but I'd recommend using an inline metafunction or a macro to do the expansion for each channel.
There was a problem hiding this comment.
that would clean up the code significantly, I will check how to do that here.
|
here is the new blending mode in action: Black Hole FX with blur enabled over a background: |
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (9)
wled00/FX_fcn.cpp (9)
1295-1305: Minor naming/usage mismatch: “unused” helpers are compiled in.
_top,_bottom,_addare marked “function currently unused” and aren’t referenced. Consider removing or markingstatic inline constexprbehind a feature flag to save a few bytes.- static uint8_t _top (uint8_t a, uint8_t b) { return a; } // function currently unused - static uint8_t _bottom (uint8_t a, uint8_t b) { return b; } // function currently unused - static uint8_t _add (uint8_t a, uint8_t b) { unsigned t = a + b; return t > 255 ? 255 : t; } // function currently unused +// #define WLED_DEBUG_BLEND_FUNCS 1 +#ifdef WLED_DEBUG_BLEND_FUNCS +static inline constexpr uint8_t _top (uint8_t a, uint8_t) { return a; } +static inline constexpr uint8_t _bottom (uint8_t, uint8_t b) { return b; } +static inline constexpr uint8_t _add (uint8_t a, uint8_t b) { unsigned t = a + b; return t > 255 ? 255 : t; } +#endif
1320-1321: Prefer an inline function over a macro for type-safety.
APPLY_BLENDrelies on outer names (tC,bC,c) and hides types. Aninlinehelper is safer and can still inline well.-#define APPLY_BLEND(op) c.r = op(tC.r, bC.r); c.g = op(tC.g, bC.g); c.b = op(tC.b, bC.b); c.w = op(tC.w, bC.w) +template<typename Op> +static inline void applyBlendOp(CRGBW& out, const CRGBW& a, const CRGBW& b, Op op) { + out.r = op(a.r, b.r); out.g = op(a.g, b.g); out.b = op(a.b, b.b); out.w = op(a.w, b.w); +}Then:
- case 3: APPLY_BLEND(_subtract); break; + case 3: applyBlendOp(c, tC, bC, _subtract); break;
1322-1348: segblend(): semantics and perf look solid; add force-inline and guard CRGBW conversions.
- Good consolidation of blend modes; stencil (32) behaves as described.
- To keep this hot path in IRAM and reduce call overhead, consider
IRAM_ATTRandalways_inline.- If
color_add()only acceptsuint32_t, relying on implicit CRGBW→u32 conversions is brittle across toolchains.-static uint32_t segblend(CRGBW tcol, CRGBW bcol, CRGBW bgcol, uint8_t blendMode) { +static inline __attribute__((always_inline)) IRAM_ATTR uint32_t segblend(CRGBW tcol, CRGBW bcol, CRGBW bgcol, uint8_t blendMode) { CRGBW tC(tcol); CRGBW bC(bcol); CRGBW c; switch (blendMode) { - case 2: return color_add(tcol, bcol, false); // Add + case 2: return color_add(tcol.color32, bcol.color32, false); // Add (explicit)Verification ask: confirm
CRGBWhas implicit conversion where used, or make all calls explicit (.color32) to avoid accidental overload picks.
1342-1346: Stencil equality: consider optional tolerance to handle palette/gamma drift.Exact 32-bit equality can fail if the top color is dithered or preprocessed. If needed later, add an optional “within N per-channel” compare guarded by a flag.
Would you like a tiny helper that compares channels within a tolerance (e.g., <=1) and is compiled out by default?
1430-1453: 2D mirroring path: index math and CCT writes look correct; avoid repeated segblend bottom fetch.You recompute
segblend(c, _pixels[idx*], ...)per mirror target (necessary because bottoms differ), but_pixels[idx*]is read twice per pixel (once for blend input, once again implicitly in color_blend). Micro-opt: read once per idx and reuse.- _pixels[indx] = color_blend(_pixels[indx], segblend(c, _pixels[indx], bgColor, blendMode), o); + uint32_t base = _pixels[indx]; + uint32_t mixed = segblend(c, base, bgColor, blendMode); + _pixels[indx] = color_blend(base, mixed, o);Same pattern for
idxMX/idxMY/idxMM.
1457-1468: Push offsets: clean and branchless; mask-based wrap is fine.Minor readability nit: unify comment (“shift canvas”) once at the switch header.
1554-1563: On/Off transition workaround still relies on global bri/briT; add a brief comment.Future readers may forget why FADE is special-cased vs others. One-line rationale would help.
1299-1319: Divide/Dodge/Burn edge cases are safe.No divide-by-zero due to guards; semantics are consistent with your chosen definitions. Consider
constexprto help constant-fold.
1322-1348: Unit test coverage for bm=32 (stencil) and regressions on bm=3 suggested.Given prior subtract-mode bug, add minimal tests (or a debug scene) to assert:
- Subtract uses bottom-minus-top across all channels.
- Stencil returns bottom when
top==bg, otherwise top.I can provide a small harness that feeds CRGBW tuples through
segblend()and checks results; want me to drop that intest/?
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
💡 Knowledge Base configuration:
- MCP integration is disabled by default for public repositories
- Jira integration is disabled by default for public repositories
- Linear integration is disabled by default for public repositories
You can enable these sources in your CodeRabbit configuration.
📒 Files selected for processing (1)
wled00/FX_fcn.cpp(9 hunks)
🧰 Additional context used
📓 Path-based instructions (1)
wled00/{*.cpp,!(html_*)*.h}
📄 CodeRabbit inference engine (.github/copilot-instructions.md)
Use spaces (2 per level) for C++ source and header files
Files:
wled00/FX_fcn.cpp
🧠 Learnings (3)
📓 Common learnings
Learnt from: DedeHai
PR: wled/WLED#4798
File: wled00/FX.cpp:7531-7533
Timestamp: 2025-08-26T11:51:21.817Z
Learning: In WLED PR #4798, DedeHai confirmed that certain gamma-related calls in FX.cpp/FX_fcn.cpp/particle systems are intentional for effect-level shaping (e.g., brightness curves, TV sim, Pride 2015 pre-mix), distinct from final output gamma. Do not flag or remove these in future reviews; add comments when feasible to clarify intent.
📚 Learning: 2025-08-26T11:51:21.817Z
Learnt from: DedeHai
PR: wled/WLED#4798
File: wled00/FX.cpp:7531-7533
Timestamp: 2025-08-26T11:51:21.817Z
Learning: In WLED PR #4798, DedeHai confirmed that certain gamma-related calls in FX.cpp/FX_fcn.cpp/particle systems are intentional for effect-level shaping (e.g., brightness curves, TV sim, Pride 2015 pre-mix), distinct from final output gamma. Do not flag or remove these in future reviews; add comments when feasible to clarify intent.
Applied to files:
wled00/FX_fcn.cpp
📚 Learning: 2025-06-07T15:58:42.579Z
Learnt from: willmmiles
PR: wled/WLED#4712
File: wled00/FX_fcn.cpp:992-993
Timestamp: 2025-06-07T15:58:42.579Z
Learning: In WLED segments, the backing pixel buffer is always allocated to the full physical segment size, but effect rendering only uses a subset when mirror or grouping are enabled. For 2D segments, this subset is vWidth() * vHeight(), and for 1D segments it's vLength(). Color processing operations like fade_out should iterate over this rendering subset, not the full allocated buffer.
Applied to files:
wled00/FX_fcn.cpp
🧬 Code graph analysis (1)
wled00/FX_fcn.cpp (1)
wled00/colors.cpp (4)
color_add(28-61)color_add(28-28)color_blend(11-21)color_blend(11-11)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (18)
- GitHub Check: wled_build / Build Environments (nodemcuv2_compat)
- GitHub Check: wled_build / Build Environments (esp32c3dev)
- GitHub Check: wled_build / Build Environments (esp32s3dev_8MB_opi)
- GitHub Check: wled_build / Build Environments (esp32_eth)
- GitHub Check: wled_build / Build Environments (esp01_1m_full_160)
- GitHub Check: wled_build / Build Environments (usermods)
- GitHub Check: wled_build / Build Environments (lolin_s2_mini)
- GitHub Check: wled_build / Build Environments (esp32_wrover)
- GitHub Check: wled_build / Build Environments (esp32s3_4M_qspi)
- GitHub Check: wled_build / Build Environments (esp8266_2m_160)
- GitHub Check: wled_build / Build Environments (esp01_1m_full_compat)
- GitHub Check: wled_build / Build Environments (esp32dev)
- GitHub Check: wled_build / Build Environments (esp8266_2m_compat)
- GitHub Check: wled_build / Build Environments (esp32s3dev_16MB_opi)
- GitHub Check: wled_build / Build Environments (nodemcuv2_160)
- GitHub Check: wled_build / Build Environments (esp01_1m_full)
- GitHub Check: wled_build / Build Environments (nodemcuv2)
- GitHub Check: wled_build / Build Environments (esp8266_2m)
🔇 Additional comments (6)
wled00/FX_fcn.cpp (6)
372-375: Good safety clamp on palette blends.Covers overflow in transition palette blending; no behavior concerns.
1350-1391: blendSegment(): clipping window math is correct; single-pixel segments fallback to FADE is a nice edge-case fix.No issues; good use of dw/dh and inverted ranges for OUTSIDE_IN.
1460-1516: Reverse/transpose gating only when needed: good for perf; group expansion bounds are correct.Use of
groupLength()and min(...) avoids overruns. Looks good.
1531-1536: 1D mirroring path mirrors before primary write; offset wrap is correct.No functional concerns; mirrors respect
offsetand segment bounds.
1575-1577: Restore global blendingStyle: good defensive reset.Prevents leakage into overlay passes.
1618-1620: Gamma guard avoids work on black; confirm realtime/gamma combo.Condition
c > 0 && !(realtimeMode && arlsDisableGammaCorrection)is correct. Verify that “pure white with CCT-only segments” still passes as >0 (it does).If you want, I can craft a quick bench to compare FPS with/without the
c > 0guard on typical scenes.
|
changed to draft, this PR needs to wait until WIP on transparency is done. |
as mentioned in #4550 there is no easy way to overlay text with the new blending modes. this addition fixes that.
I also made some slight optimisations to
blendSegment()and found that the compiler is very unpredictable. I assume that is mostly due to pipelining and core register use. In general, replacing bit-field accessed variables with an alias or a pre-computed value is faster, but here it is not always the case. I tried different variations and this was the best performing one. Replacing any other repeated calculations inblendSegment()does not improve performance.while flash size increases by a few hundred bytes, FPS improve by 3-5% depending on FX and layering used.
Summary by CodeRabbit
New Features
Bug Fixes
Notes