perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
#include "FontDecompressor.h"
|
|
|
|
|
|
2026-03-12 07:05:46 +11:00
|
|
|
#include <Arduino.h>
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
#include <Logging.h>
|
2026-03-12 07:05:46 +11:00
|
|
|
#include <Utf8.h>
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
|
|
|
|
|
#include <cstdlib>
|
|
|
|
|
|
2026-03-12 07:05:46 +11:00
|
|
|
FontDecompressor::~FontDecompressor() { deinit(); }
|
|
|
|
|
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
bool FontDecompressor::init() {
|
|
|
|
|
clearCache();
|
|
|
|
|
return true;
|
|
|
|
|
}
|
|
|
|
|
|
2026-03-12 07:05:46 +11:00
|
|
|
void FontDecompressor::deinit() {
|
|
|
|
|
freePageBuffer();
|
|
|
|
|
freeHotGroup();
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void FontDecompressor::clearCache() {
|
2026-03-12 07:05:46 +11:00
|
|
|
freePageBuffer();
|
|
|
|
|
freeHotGroup();
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void FontDecompressor::freePageBuffer() {
|
2026-03-22 04:10:41 +11:00
|
|
|
for (uint8_t s = 0; s < pageSlotCount; s++) {
|
|
|
|
|
free(pageSlots[s].buffer);
|
|
|
|
|
free(pageSlots[s].glyphs);
|
|
|
|
|
pageSlots[s] = {};
|
|
|
|
|
}
|
|
|
|
|
pageSlotCount = 0;
|
2026-03-12 07:05:46 +11:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void FontDecompressor::freeHotGroup() {
|
|
|
|
|
hotGroup.clear();
|
|
|
|
|
hotGroup.shrink_to_fit();
|
|
|
|
|
hotGroupFont = nullptr;
|
|
|
|
|
hotGroupIndex = UINT16_MAX;
|
|
|
|
|
hotGlyphBuf.clear();
|
|
|
|
|
hotGlyphBuf.shrink_to_fit();
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
}
|
|
|
|
|
|
2026-03-12 07:05:46 +11:00
|
|
|
uint16_t FontDecompressor::getGroupIndex(const EpdFontData* fontData, uint32_t glyphIndex) {
|
|
|
|
|
// O(1) path for frequency-grouped fonts with glyphToGroup mapping
|
|
|
|
|
if (fontData->glyphToGroup != nullptr) {
|
|
|
|
|
return fontData->glyphToGroup[glyphIndex];
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Contiguous-group fonts: linear scan
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
for (uint16_t i = 0; i < fontData->groupCount; i++) {
|
2026-03-12 07:05:46 +11:00
|
|
|
uint32_t first = fontData->groups[i].firstGlyphIndex;
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
if (glyphIndex >= first && glyphIndex < first + fontData->groups[i].glyphCount) {
|
|
|
|
|
return i;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
return fontData->groupCount; // sentinel = not found
|
|
|
|
|
}
|
|
|
|
|
|
2026-03-12 07:05:46 +11:00
|
|
|
bool FontDecompressor::decompressGroup(const EpdFontData* fontData, uint16_t groupIndex, uint8_t* outBuf,
|
|
|
|
|
uint32_t outSize) {
|
|
|
|
|
const EpdFontGroup& group = fontData->groups[groupIndex];
|
|
|
|
|
|
|
|
|
|
const uint32_t tDecomp = millis();
|
|
|
|
|
inflateReader.init(false);
|
|
|
|
|
inflateReader.setSource(&fontData->bitmap[group.compressedOffset], group.compressedSize);
|
|
|
|
|
if (!inflateReader.read(outBuf, outSize)) {
|
|
|
|
|
stats.decompressTimeMs += millis() - tDecomp;
|
|
|
|
|
LOG_ERR("FDC", "Decompression failed for group %u", groupIndex);
|
|
|
|
|
return false;
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
}
|
2026-03-12 07:05:46 +11:00
|
|
|
stats.decompressTimeMs += millis() - tDecomp;
|
|
|
|
|
return true;
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
}
|
|
|
|
|
|
2026-03-12 07:05:46 +11:00
|
|
|
// --- Byte-aligned helpers ---
|
|
|
|
|
|
|
|
|
|
uint32_t FontDecompressor::getAlignedOffset(const EpdFontData* fontData, uint16_t groupIndex, uint32_t glyphIndex) {
|
|
|
|
|
uint32_t offset = 0;
|
|
|
|
|
|
|
|
|
|
auto accumGlyph = [&](const EpdGlyph& g) {
|
|
|
|
|
if (g.width > 0 && g.height > 0) {
|
|
|
|
|
offset += ((g.width + 3) / 4) * g.height;
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
}
|
2026-03-12 07:05:46 +11:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
if (fontData->glyphToGroup) {
|
|
|
|
|
// Frequency-grouped: scan glyphs before glyphIndex that belong to this group
|
|
|
|
|
for (uint32_t i = 0; i < glyphIndex; i++) {
|
|
|
|
|
if (fontData->glyphToGroup[i] == groupIndex) {
|
|
|
|
|
accumGlyph(fontData->glyph[i]);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
} else {
|
|
|
|
|
// Contiguous-group: sum aligned sizes of preceding glyphs in the group
|
|
|
|
|
const EpdFontGroup& group = fontData->groups[groupIndex];
|
|
|
|
|
for (uint32_t i = group.firstGlyphIndex; i < glyphIndex; i++) {
|
|
|
|
|
accumGlyph(fontData->glyph[i]);
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2026-03-12 07:05:46 +11:00
|
|
|
return offset;
|
|
|
|
|
}
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
|
2026-03-12 07:05:46 +11:00
|
|
|
void FontDecompressor::compactSingleGlyph(const uint8_t* alignedSrc, uint8_t* packedDst, uint8_t width,
|
|
|
|
|
uint8_t height) {
|
|
|
|
|
if (width == 0 || height == 0) return;
|
|
|
|
|
const uint32_t rowStride = (width + 3) / 4;
|
|
|
|
|
if (width % 4 == 0) {
|
|
|
|
|
memcpy(packedDst, alignedSrc, rowStride * height);
|
|
|
|
|
return;
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
}
|
2026-03-12 07:05:46 +11:00
|
|
|
uint8_t outByte = 0, outBits = 0;
|
|
|
|
|
uint32_t writeIdx = 0;
|
|
|
|
|
for (uint8_t y = 0; y < height; y++) {
|
|
|
|
|
for (uint8_t x = 0; x < width; x++) {
|
|
|
|
|
outByte = (outByte << 2) | ((alignedSrc[y * rowStride + x / 4] >> ((3 - (x % 4)) * 2)) & 0x3);
|
|
|
|
|
outBits += 2;
|
|
|
|
|
if (outBits == 8) {
|
|
|
|
|
packedDst[writeIdx++] = outByte;
|
|
|
|
|
outByte = 0;
|
|
|
|
|
outBits = 0;
|
|
|
|
|
}
|
|
|
|
|
}
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
}
|
2026-03-12 07:05:46 +11:00
|
|
|
if (outBits > 0) packedDst[writeIdx] = outByte << (8 - outBits);
|
|
|
|
|
}
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
|
2026-03-12 07:05:46 +11:00
|
|
|
// --- getBitmap: page buffer → hot group → decompress ---
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
|
2026-03-12 07:05:46 +11:00
|
|
|
const uint8_t* FontDecompressor::getBitmap(const EpdFontData* fontData, const EpdGlyph* glyph, uint32_t glyphIndex) {
|
|
|
|
|
const uint32_t tStart = micros();
|
|
|
|
|
stats.getBitmapCalls++;
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
|
|
|
|
|
if (!fontData->groups || fontData->groupCount == 0) {
|
2026-03-12 07:05:46 +11:00
|
|
|
stats.getBitmapTimeUs += micros() - tStart;
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
return &fontData->bitmap[glyph->dataOffset];
|
|
|
|
|
}
|
|
|
|
|
|
2026-03-22 04:10:41 +11:00
|
|
|
// Check page buffer slots (populated by prewarmCache — one slot per font style)
|
|
|
|
|
for (uint8_t s = 0; s < pageSlotCount; s++) {
|
|
|
|
|
const auto& slot = pageSlots[s];
|
|
|
|
|
if (slot.fontData != fontData || slot.glyphCount == 0) continue;
|
|
|
|
|
|
|
|
|
|
int left = 0, right = slot.glyphCount - 1;
|
2026-03-12 07:05:46 +11:00
|
|
|
while (left <= right) {
|
|
|
|
|
int mid = left + (right - left) / 2;
|
2026-03-22 04:10:41 +11:00
|
|
|
if (slot.glyphs[mid].glyphIndex == glyphIndex) {
|
|
|
|
|
if (slot.glyphs[mid].bufferOffset != UINT32_MAX) {
|
2026-03-12 07:05:46 +11:00
|
|
|
stats.cacheHits++;
|
|
|
|
|
stats.getBitmapTimeUs += micros() - tStart;
|
2026-03-22 04:10:41 +11:00
|
|
|
return &slot.buffer[slot.glyphs[mid].bufferOffset];
|
2026-03-12 07:05:46 +11:00
|
|
|
}
|
|
|
|
|
break; // Not extracted during prewarm; fall through to hot-group path
|
|
|
|
|
}
|
2026-03-22 04:10:41 +11:00
|
|
|
if (slot.glyphs[mid].glyphIndex < glyphIndex)
|
2026-03-12 07:05:46 +11:00
|
|
|
left = mid + 1;
|
|
|
|
|
else
|
|
|
|
|
right = mid - 1;
|
|
|
|
|
}
|
2026-03-22 04:10:41 +11:00
|
|
|
break; // Found the right slot but glyph wasn't in it; don't check other slots
|
2026-03-12 07:05:46 +11:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Fallback: hot group slot
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
uint16_t groupIndex = getGroupIndex(fontData, glyphIndex);
|
|
|
|
|
if (groupIndex >= fontData->groupCount) {
|
|
|
|
|
LOG_ERR("FDC", "Glyph %u not found in any group", glyphIndex);
|
2026-03-12 07:05:46 +11:00
|
|
|
stats.getBitmapTimeUs += micros() - tStart;
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
return nullptr;
|
|
|
|
|
}
|
|
|
|
|
|
2026-03-12 07:05:46 +11:00
|
|
|
// Check if hot group already has this group decompressed — if not, decompress it
|
|
|
|
|
if (!(!hotGroup.empty() && hotGroupFont == fontData && hotGroupIndex == groupIndex)) {
|
|
|
|
|
stats.cacheMisses++;
|
|
|
|
|
const EpdFontGroup& group = fontData->groups[groupIndex];
|
|
|
|
|
|
|
|
|
|
hotGroup.resize(group.uncompressedSize);
|
|
|
|
|
if (hotGroup.empty()) {
|
|
|
|
|
LOG_ERR("FDC", "Failed to allocate %u bytes for hot group %u", group.uncompressedSize, groupIndex);
|
|
|
|
|
hotGroupFont = nullptr;
|
|
|
|
|
hotGroupIndex = UINT16_MAX;
|
|
|
|
|
stats.getBitmapTimeUs += micros() - tStart;
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
return nullptr;
|
|
|
|
|
}
|
2026-03-12 07:05:46 +11:00
|
|
|
|
|
|
|
|
if (!decompressGroup(fontData, groupIndex, hotGroup.data(), group.uncompressedSize)) {
|
|
|
|
|
hotGroup.clear();
|
|
|
|
|
hotGroup.shrink_to_fit();
|
|
|
|
|
hotGroupFont = nullptr;
|
|
|
|
|
hotGroupIndex = UINT16_MAX;
|
|
|
|
|
stats.getBitmapTimeUs += micros() - tStart;
|
|
|
|
|
return nullptr;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
hotGroupFont = fontData;
|
|
|
|
|
hotGroupIndex = groupIndex;
|
|
|
|
|
stats.hotGroupBytes = group.uncompressedSize;
|
|
|
|
|
} else {
|
|
|
|
|
stats.cacheHits++;
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
}
|
|
|
|
|
|
2026-03-12 07:05:46 +11:00
|
|
|
// Compact just the requested glyph from byte-aligned data into scratch buffer
|
|
|
|
|
if (glyph->dataLength > hotGlyphBuf.size()) {
|
|
|
|
|
hotGlyphBuf.resize(glyph->dataLength);
|
|
|
|
|
}
|
|
|
|
|
if (hotGlyphBuf.empty()) {
|
|
|
|
|
stats.getBitmapTimeUs += micros() - tStart;
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
return nullptr;
|
|
|
|
|
}
|
|
|
|
|
|
2026-03-12 07:05:46 +11:00
|
|
|
uint32_t alignedOff = getAlignedOffset(fontData, groupIndex, glyphIndex);
|
|
|
|
|
compactSingleGlyph(&hotGroup[alignedOff], hotGlyphBuf.data(), glyph->width, glyph->height);
|
|
|
|
|
stats.getBitmapTimeUs += micros() - tStart;
|
|
|
|
|
return hotGlyphBuf.data();
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// --- Prewarm: pre-decompress glyph bitmaps for a page of text ---
|
|
|
|
|
|
|
|
|
|
int32_t FontDecompressor::findGlyphIndex(const EpdFontData* fontData, uint32_t codepoint) {
|
|
|
|
|
const EpdUnicodeInterval* intervals = fontData->intervals;
|
|
|
|
|
const int count = fontData->intervalCount;
|
|
|
|
|
|
|
|
|
|
if (count == 0) return -1;
|
|
|
|
|
|
|
|
|
|
// Binary search
|
|
|
|
|
int left = 0;
|
|
|
|
|
int right = count - 1;
|
|
|
|
|
|
|
|
|
|
while (left <= right) {
|
|
|
|
|
const int mid = left + (right - left) / 2;
|
|
|
|
|
const EpdUnicodeInterval* interval = &intervals[mid];
|
|
|
|
|
|
|
|
|
|
if (codepoint < interval->first) {
|
|
|
|
|
right = mid - 1;
|
|
|
|
|
} else if (codepoint > interval->last) {
|
|
|
|
|
left = mid + 1;
|
|
|
|
|
} else {
|
|
|
|
|
return static_cast<int32_t>(interval->offset + (codepoint - interval->first));
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return -1;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
int FontDecompressor::prewarmCache(const EpdFontData* fontData, const char* utf8Text) {
|
|
|
|
|
if (!fontData || !fontData->groups || !utf8Text) return 0;
|
|
|
|
|
|
2026-03-22 04:10:41 +11:00
|
|
|
// Allocate the next available slot (caller must call freePageBuffer/clearCache to reset)
|
|
|
|
|
if (pageSlotCount >= MAX_PAGE_SLOTS) {
|
|
|
|
|
LOG_ERR("FDC", "All %u page buffer slots full, cannot prewarm fontData=%p", MAX_PAGE_SLOTS, (void*)fontData);
|
|
|
|
|
return -1;
|
|
|
|
|
}
|
|
|
|
|
PageSlot& slot = pageSlots[pageSlotCount];
|
|
|
|
|
|
2026-03-12 07:05:46 +11:00
|
|
|
// Step 1: Collect unique glyph indices needed for this page
|
|
|
|
|
uint32_t neededGlyphs[MAX_PAGE_GLYPHS];
|
|
|
|
|
uint16_t glyphCount = 0;
|
|
|
|
|
bool glyphCapWarned = false;
|
|
|
|
|
|
|
|
|
|
const unsigned char* p = reinterpret_cast<const unsigned char*>(utf8Text);
|
|
|
|
|
while (*p) {
|
|
|
|
|
uint32_t cp = utf8NextCodepoint(&p);
|
|
|
|
|
if (cp == 0) break;
|
|
|
|
|
|
|
|
|
|
int32_t glyphIdx = findGlyphIndex(fontData, cp);
|
|
|
|
|
if (glyphIdx < 0) continue;
|
|
|
|
|
|
|
|
|
|
// Deduplicate
|
|
|
|
|
bool found = false;
|
|
|
|
|
for (uint16_t i = 0; i < glyphCount; i++) {
|
|
|
|
|
if (neededGlyphs[i] == static_cast<uint32_t>(glyphIdx)) {
|
|
|
|
|
found = true;
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
if (!found) {
|
|
|
|
|
if (glyphCount < MAX_PAGE_GLYPHS) {
|
|
|
|
|
neededGlyphs[glyphCount++] = static_cast<uint32_t>(glyphIdx);
|
|
|
|
|
} else if (!glyphCapWarned) {
|
|
|
|
|
LOG_DBG("FDC", "Glyph cap (%u) reached during prewarm; excess glyphs will use hot-group fallback",
|
|
|
|
|
MAX_PAGE_GLYPHS);
|
|
|
|
|
glyphCapWarned = true;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (glyphCount == 0) return 0;
|
|
|
|
|
|
|
|
|
|
// Step 2: Compute total buffer size and collect unique groups
|
|
|
|
|
uint32_t totalBytes = 0;
|
|
|
|
|
uint16_t neededGroups[128];
|
|
|
|
|
uint8_t groupCount = 0;
|
|
|
|
|
bool groupCapWarned = false;
|
|
|
|
|
|
|
|
|
|
for (uint16_t i = 0; i < glyphCount; i++) {
|
|
|
|
|
totalBytes += fontData->glyph[neededGlyphs[i]].dataLength;
|
|
|
|
|
uint16_t gi = getGroupIndex(fontData, neededGlyphs[i]);
|
|
|
|
|
bool found = false;
|
|
|
|
|
for (uint8_t j = 0; j < groupCount; j++) {
|
|
|
|
|
if (neededGroups[j] == gi) {
|
|
|
|
|
found = true;
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
if (!found) {
|
|
|
|
|
if (groupCount < 128) {
|
|
|
|
|
neededGroups[groupCount++] = gi;
|
|
|
|
|
} else if (!groupCapWarned) {
|
|
|
|
|
LOG_DBG("FDC", "Group cap (128) reached during prewarm; some groups will use hot-group fallback");
|
|
|
|
|
groupCapWarned = true;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
stats.uniqueGroupsAccessed = groupCount;
|
|
|
|
|
|
2026-03-22 04:10:41 +11:00
|
|
|
// Step 3: Allocate page buffer and lookup table for this slot
|
|
|
|
|
slot.buffer = static_cast<uint8_t*>(malloc(totalBytes));
|
|
|
|
|
slot.glyphs = static_cast<PageGlyphEntry*>(malloc(glyphCount * sizeof(PageGlyphEntry)));
|
|
|
|
|
if (!slot.buffer || !slot.glyphs) {
|
2026-03-12 07:05:46 +11:00
|
|
|
LOG_ERR("FDC", "Failed to allocate page buffer (%u bytes, %u glyphs)", totalBytes, glyphCount);
|
2026-03-22 04:10:41 +11:00
|
|
|
free(slot.buffer);
|
|
|
|
|
free(slot.glyphs);
|
|
|
|
|
slot = {};
|
2026-03-12 07:05:46 +11:00
|
|
|
return glyphCount;
|
|
|
|
|
}
|
2026-03-22 04:10:41 +11:00
|
|
|
stats.pageBufferBytes += totalBytes;
|
|
|
|
|
stats.pageGlyphsBytes += glyphCount * sizeof(PageGlyphEntry);
|
2026-03-12 07:05:46 +11:00
|
|
|
|
2026-03-22 04:10:41 +11:00
|
|
|
slot.fontData = fontData;
|
|
|
|
|
slot.glyphCount = glyphCount;
|
|
|
|
|
pageSlotCount++;
|
2026-03-12 07:05:46 +11:00
|
|
|
|
|
|
|
|
// Initialize lookup entries (bufferOffset = UINT32_MAX means not yet extracted)
|
|
|
|
|
for (uint16_t i = 0; i < glyphCount; i++) {
|
2026-03-22 04:10:41 +11:00
|
|
|
slot.glyphs[i] = {neededGlyphs[i], UINT32_MAX, 0};
|
2026-03-12 07:05:46 +11:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Sort by glyphIndex for binary search in getBitmap()
|
|
|
|
|
for (uint16_t i = 1; i < glyphCount; i++) {
|
2026-03-22 04:10:41 +11:00
|
|
|
PageGlyphEntry key = slot.glyphs[i];
|
2026-03-12 07:05:46 +11:00
|
|
|
int j = i - 1;
|
2026-03-22 04:10:41 +11:00
|
|
|
while (j >= 0 && slot.glyphs[j].glyphIndex > key.glyphIndex) {
|
|
|
|
|
slot.glyphs[j + 1] = slot.glyphs[j];
|
2026-03-12 07:05:46 +11:00
|
|
|
j--;
|
|
|
|
|
}
|
2026-03-22 04:10:41 +11:00
|
|
|
slot.glyphs[j + 1] = key;
|
2026-03-12 07:05:46 +11:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Step 3b: Pre-scan to compute each needed glyph's byte-aligned offset within its group.
|
|
|
|
|
// This avoids recomputing aligned offsets per group during extraction in step 4.
|
|
|
|
|
uint32_t groupAlignedTracker[128] = {}; // running byte-aligned offset for each needed group
|
|
|
|
|
|
|
|
|
|
if (fontData->glyphToGroup) {
|
|
|
|
|
// Frequency-grouped: single O(totalGlyphs) pass through glyphToGroup
|
|
|
|
|
const auto& lastInterval = fontData->intervals[fontData->intervalCount - 1];
|
|
|
|
|
const uint32_t totalGlyphs = lastInterval.offset + (lastInterval.last - lastInterval.first + 1);
|
|
|
|
|
|
|
|
|
|
for (uint32_t i = 0; i < totalGlyphs; i++) {
|
|
|
|
|
const uint16_t gi = fontData->glyphToGroup[i];
|
|
|
|
|
// Find this glyph's group position in neededGroups
|
|
|
|
|
uint8_t gpPos = groupCount;
|
|
|
|
|
for (uint8_t j = 0; j < groupCount; j++) {
|
|
|
|
|
if (neededGroups[j] == gi) {
|
|
|
|
|
gpPos = j;
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
if (gpPos == groupCount) continue; // not a needed group
|
|
|
|
|
|
|
|
|
|
const EpdGlyph& glyph = fontData->glyph[i];
|
|
|
|
|
|
2026-03-22 04:10:41 +11:00
|
|
|
// Binary search in sorted slot.glyphs to find if glyph i is needed
|
|
|
|
|
int left = 0, right = (int)slot.glyphCount - 1;
|
2026-03-12 07:05:46 +11:00
|
|
|
while (left <= right) {
|
|
|
|
|
const int mid = left + (right - left) / 2;
|
2026-03-22 04:10:41 +11:00
|
|
|
if (slot.glyphs[mid].glyphIndex == i) {
|
|
|
|
|
slot.glyphs[mid].alignedOffset = groupAlignedTracker[gpPos];
|
2026-03-12 07:05:46 +11:00
|
|
|
break;
|
|
|
|
|
}
|
2026-03-22 04:10:41 +11:00
|
|
|
if (slot.glyphs[mid].glyphIndex < i)
|
2026-03-12 07:05:46 +11:00
|
|
|
left = mid + 1;
|
|
|
|
|
else
|
|
|
|
|
right = mid - 1;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (glyph.width > 0 && glyph.height > 0) {
|
|
|
|
|
groupAlignedTracker[gpPos] += ((glyph.width + 3) / 4) * glyph.height;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
} else {
|
|
|
|
|
// Contiguous-group: iterate each needed group's glyphs directly
|
|
|
|
|
for (uint8_t g = 0; g < groupCount; g++) {
|
|
|
|
|
const EpdFontGroup& group = fontData->groups[neededGroups[g]];
|
|
|
|
|
uint32_t alignedOff = 0;
|
|
|
|
|
for (uint16_t j = 0; j < group.glyphCount; j++) {
|
|
|
|
|
const uint32_t glyphI = group.firstGlyphIndex + j;
|
|
|
|
|
const EpdGlyph& glyph = fontData->glyph[glyphI];
|
|
|
|
|
|
2026-03-22 04:10:41 +11:00
|
|
|
int left = 0, right = (int)slot.glyphCount - 1;
|
2026-03-12 07:05:46 +11:00
|
|
|
while (left <= right) {
|
|
|
|
|
const int mid = left + (right - left) / 2;
|
2026-03-22 04:10:41 +11:00
|
|
|
if (slot.glyphs[mid].glyphIndex == glyphI) {
|
|
|
|
|
slot.glyphs[mid].alignedOffset = alignedOff;
|
2026-03-12 07:05:46 +11:00
|
|
|
break;
|
|
|
|
|
}
|
2026-03-22 04:10:41 +11:00
|
|
|
if (slot.glyphs[mid].glyphIndex < glyphI)
|
2026-03-12 07:05:46 +11:00
|
|
|
left = mid + 1;
|
|
|
|
|
else
|
|
|
|
|
right = mid - 1;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (glyph.width > 0 && glyph.height > 0) {
|
|
|
|
|
alignedOff += ((glyph.width + 3) / 4) * glyph.height;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Step 4: For each unique group, decompress to temp buffer and extract needed glyphs
|
|
|
|
|
uint32_t writeOffset = 0;
|
|
|
|
|
int missed = 0;
|
|
|
|
|
|
|
|
|
|
for (uint8_t g = 0; g < groupCount; g++) {
|
|
|
|
|
uint16_t groupIdx = neededGroups[g];
|
|
|
|
|
const EpdFontGroup& group = fontData->groups[groupIdx];
|
|
|
|
|
|
|
|
|
|
auto* tempBuf = static_cast<uint8_t*>(malloc(group.uncompressedSize));
|
|
|
|
|
if (!tempBuf) {
|
|
|
|
|
LOG_ERR("FDC", "Failed to allocate temp buffer (%u bytes) for group %u", group.uncompressedSize, groupIdx);
|
|
|
|
|
missed++;
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
if (group.uncompressedSize > stats.peakTempBytes) {
|
|
|
|
|
stats.peakTempBytes = group.uncompressedSize;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (!decompressGroup(fontData, groupIdx, tempBuf, group.uncompressedSize)) {
|
|
|
|
|
free(tempBuf);
|
|
|
|
|
missed++;
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Extract needed glyphs directly from the byte-aligned temp buffer, compacting on the fly.
|
|
|
|
|
// alignedOffset was pre-computed in step 3b — no full-group compact scan needed.
|
2026-03-22 04:10:41 +11:00
|
|
|
for (uint16_t i = 0; i < slot.glyphCount; i++) {
|
|
|
|
|
if (slot.glyphs[i].bufferOffset != UINT32_MAX) continue; // already extracted
|
|
|
|
|
if (getGroupIndex(fontData, slot.glyphs[i].glyphIndex) != groupIdx) continue;
|
2026-03-12 07:05:46 +11:00
|
|
|
|
2026-03-22 04:10:41 +11:00
|
|
|
const EpdGlyph& glyph = fontData->glyph[slot.glyphs[i].glyphIndex];
|
|
|
|
|
compactSingleGlyph(&tempBuf[slot.glyphs[i].alignedOffset], &slot.buffer[writeOffset], glyph.width, glyph.height);
|
|
|
|
|
slot.glyphs[i].bufferOffset = writeOffset;
|
2026-03-12 07:05:46 +11:00
|
|
|
writeOffset += glyph.dataLength;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
free(tempBuf);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
LOG_DBG("FDC", "Prewarm: %u glyphs in %u bytes from %u groups (%d missed)", glyphCount, writeOffset, groupCount,
|
|
|
|
|
missed);
|
|
|
|
|
|
|
|
|
|
return missed;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// --- Stats ---
|
|
|
|
|
|
|
|
|
|
void FontDecompressor::resetStats() { stats = Stats{}; }
|
|
|
|
|
|
|
|
|
|
void FontDecompressor::logStats(const char* label) {
|
|
|
|
|
const uint32_t total = stats.cacheHits + stats.cacheMisses;
|
|
|
|
|
LOG_DBG("FDC", "[%s] hits=%lu misses=%lu (%.1f%% hit rate)", label, stats.cacheHits, stats.cacheMisses,
|
|
|
|
|
total > 0 ? 100.0f * stats.cacheHits / total : 0.0f);
|
|
|
|
|
LOG_DBG("FDC", "[%s] decompress=%lums groups_accessed=%u", label, stats.decompressTimeMs, stats.uniqueGroupsAccessed);
|
|
|
|
|
LOG_DBG("FDC", "[%s] mem: pageBuf=%lu pageGlyphs=%lu hotGroup=%lu peakTemp=%lu", label, stats.pageBufferBytes,
|
|
|
|
|
stats.pageGlyphsBytes, stats.hotGroupBytes, stats.peakTempBytes);
|
|
|
|
|
if (stats.getBitmapCalls > 0) {
|
|
|
|
|
LOG_DBG("FDC", "[%s] getBitmap: %lu calls, %luus total, %luus/call avg", label, stats.getBitmapCalls,
|
|
|
|
|
stats.getBitmapTimeUs, stats.getBitmapTimeUs / stats.getBitmapCalls);
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
}
|
2026-03-12 07:05:46 +11:00
|
|
|
resetStats();
|
perf: Reduce overall flash usage by 30.7% by compressing built-in fonts (#831)
## Summary
**What is the goal of this PR?**
Compress reader font bitmaps to reduce flash usage by 30.7%.
**What changes are included?**
- New `EpdFontGroup` struct and extended `EpdFontData` with
`groups`/`groupCount` fields
- `--compress` flag in `fontconvert.py`: groups glyphs (ASCII base group
+ groups of 8) and compresses each with raw DEFLATE
- `FontDecompressor` class with 4-slot LRU cache for on-demand
decompression during rendering
- `GfxRenderer` transparently routes bitmap access through
`getGlyphBitmap()` (compressed or direct flash)
- Uses `uzlib` for decompression with minimal heap overhead.
- 48 reader fonts (Bookerly, NotoSans 12-18pt, OpenDyslexic) regenerated
with compression; 5 UI fonts unchanged
- Round-trip verification script (`verify_compression.py`) runs as part
of font generation
## Additional Context
## Flash & RAM
| | baseline | font-compression | Difference |
|--|--------|-----------------|------------|
| Flash (ELF) | 6,302,476 B (96.2%) | 4,365,022 B (66.6%) | -1,937,454 B
(-30.7%) |
| firmware.bin | 6,468,192 B | 4,531,008 B | -1,937,184 B (-29.9%) |
| RAM | 101,700 B (31.0%) | 103,076 B (31.5%) | +1,376 B (+0.5%) |
## Script-Based Grouping (Cold Cache)
Comparison of uncompressed baseline vs script-based group compression
(4-slot LRU cache, cleared each page). Glyphs are grouped by Unicode
block (ASCII, Latin-1, Latin Extended-A, Combining Marks, Cyrillic,
General Punctuation, etc.) instead of sequential groups of 8.
### Render Time
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Median** | 414.9 ms | 431.6 ms | +16.7 ms (+4.0%) |
| **Pages** | 37 | 37 | |
### Memory Usage
| | Baseline | Compressed (cold cache) | Difference |
|---|---|---|---|
| **Heap free (median)** | 187.0 KB | 176.3 KB | -10.7 KB |
| **Heap free (min)** | 186.0 KB | 166.5 KB | -19.5 KB |
| **Largest block (median)** | 148.0 KB | 128.0 KB | -20.0 KB |
| **Largest block (min)** | 148.0 KB | 120.0 KB | -28.0 KB |
### Cache Effectiveness
| | Misses/page | Hit rate |
|---|---|---|
| **Compressed (cold cache)** | 2.1 | 99.85% |
------
### AI Usage
While CrossPoint doesn't have restrictions on AI tools in contributing,
please be transparent about their usage as it
helps set the right context for reviewers.
Did you use AI tools to help write this code? _**YES**_
Implementation was done by Claude Code (Opus 4.6) based on a plan
developed collaboratively. All generated font headers were verified with
an automated round-trip decompression test. The firmware was compiled
successfully but has not yet been tested on-device.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-19 20:30:15 +11:00
|
|
|
}
|