Adopt CSS-style drop shadow mask #146
Conversation
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
| #define SHADOW_LUT_Y_LEN (CONFIG_VERTICAL_OFFSET + CONFIG_SHADOW_FADE_TAIL) | ||
|
|
||
| /* Fast lookup: 17-entry approximation of (1 - t^2)^2 * 0.92 + 0.08. */ | ||
| static inline twin_fixed_t shadow_gaussian_weight(twin_fixed_t t) |
There was a problem hiding this comment.
It seems that the shadow_gaussian_weight() function has the same effect as twin_stack_blur(). Removing the call to twin_stack_blur() still produces a gradient color effect.
There was a problem hiding this comment.
It seems that the shadow_gaussian_weight() function has the same effect as twin_stack_blur().
No, these two functions have completely different purposes and effects.
The purpose of twin_stack_blur:
- Blurs an existing image in-place
- Uses the Stack Blur algorithm (2-pass separable convolution)
- Horizontal scan + vertical scan
- Creates a temporary pixmap for two-directional processing
Complexity: O(2 × width × height × k)
The purpose of shadow_gaussian_weight:
- Looks up a single weight value from a precomputed LUT
- Input: t ∈ [0, 1] (fixed-point format)
- Output: Gaussian weight (16.16 fixed-point)
- Uses formula: (1 - t²)² × 0.92 + 0.08
Complexity: O(1) lookup
Comparisons:
twin_stack_bluris like taking a photograph and applying a Photoshop blur filter to it. You're processing what's already there.shadow_gaussian_weightis like looking up paint opacity values from a chart to paint a gradient from scratch. You're creating something new.
| twin_coord_t bottom_start = win_height; | ||
| if (bottom_start < 0) | ||
| bottom_start = 0; | ||
| twin_coord_t bottom_end = win_height + bottom_extent; |
There was a problem hiding this comment.
twin_coord_t win_height = shadow->height - shadow->window->shadow_y;
twin_coord_t bottom_extent = shadow->height - win_height;
Therefore, bottom_end == shadow->height
| if (bottom_start < 0) | ||
| bottom_start = 0; | ||
| twin_coord_t bottom_end = win_height + bottom_extent; | ||
| if (bottom_end > shadow->height) |
There was a problem hiding this comment.
There is no need to compare.
There was a problem hiding this comment.
There is no need to compare.
Scenario where bounds check matters:
CONFIG_VERTICAL_OFFSET = 6, FADE_TAIL = 2
∴ lut_y_len = 8
Case 1: shadow_y = 10
→ bottom_extent limited to 8
→ bottom_end = win_height + 8
→ Need bounds check if win_height + 8 > shadow->height
Case 2: shadow_y = 20 (large window)
→ bottom_extent limited to 8
→ bottom_end = win_height + 8 (definitely < shadow->height)
Bounds check is REQUIRED because bottom_extent gets clamped to LUT size.
| if (bottom_width <= 0) | ||
| continue; | ||
|
|
||
| twin_coord_t dist_y = y - win_height; |
There was a problem hiding this comment.
twin_coord_t dist_y = y - bottom_start; this is clearer.
There was a problem hiding this comment.
twin_coord_t dist_y = y - bottom_start;this is clearer.
Current implementation is correct for LUT indexing semantics.
- alpha_lut_y is built with indices [0, y_offset)
- LUT entry 0 = maximum alpha (at window edge)
- LUT entry y_offset-1 = minimum alpha (at shadow fade end)
- dist_y must represent distance from window edge (win_height)
- NOT distance from loop start (bottom_start)
|
Only need to keep one variable for the same function. |
This commit generates shadow_gaussian lookup table during the build, so the renderer stops depending on runtime math. It replaces the stack-blur border writes with a lookup-table renderer that matches CSS behavior and stays efficient. For performance considerations, it caches the vertical and bottom falloff tables and rely on memset() or pointer writes to trim per-frame work. It exposes a small configurable fade tail and update defaults plus the API, so it aligns with CSS box-shadow semantics.
This commit generates shadow_gaussian lookup table during the build, so the renderer stops depending on runtime math. It replaces the stack-blur border writes with a lookup-table renderer that matches CSS behavior and stays efficient.
For performance considerations, it caches the vertical and bottom falloff tables and rely on memset() or pointer writes to trim per-frame work. It exposes a small configurable fade tail and update defaults plus the API, so it aligns with CSS box-shadow semantics.
Summary by cubic
Switches drop shadow rendering to a CSS-style box-shadow mask using a precomputed Gaussian LUT. Improves visual match and speeds up rendering while simplifying the shadow API.
New Features
Migration