Implement sparse css cache plus dynamic banding for AA

lib/Epub/Epub.cpp

@@ -454,6 +454,11 @@ void Epub::parseCssFiles() const {
   }
 
   // No cache yet - parse CSS files
+  if (!cssParser->beginCacheCompile()) {
+    LOG_ERR("EBP", "Failed to start CSS compile pipeline");
+    return;
+  }
+
   for (const auto& cssPath : cssFiles) {
     LOG_DBG("EBP", "Parsing CSS file: %s", cssPath.c_str());
 
@@ -496,16 +501,17 @@ void Epub::parseCssFiles() const {
       Storage.remove(tmpCssPath.c_str());
       continue;
     }
-    cssParser->loadFromStream(tempCssFile);
+    if (!cssParser->appendCompiledFromStream(tempCssFile)) {
+      LOG_ERR("EBP", "Failed to compile CSS file: %s", cssPath.c_str());
+    }
     tempCssFile.close();
     Storage.remove(tmpCssPath.c_str());
   }
 
-  // Save to cache for next time
+  // Finalize compact cache for next time.
-  if (!cssParser->saveToCache()) {
+  if (!cssParser->endCacheCompile()) {
-    LOG_ERR("EBP", "Failed to save CSS rules to cache");
+    LOG_ERR("EBP", "Failed to finalize CSS rules cache");
   }
-  cssParser->clear();
 
   LOG_DBG("EBP", "Loaded %zu CSS style rules from %zu files", cssParser->ruleCount(), cssFiles.size());
 }

lib/Epub/Epub/Section.cpp

@@ -3,6 +3,8 @@
 #include <HalStorage.h>
 #include <Logging.h>
 #include <Serialization.h>
+#include <esp_heap_caps.h>
+#include <esp_system.h>
 
 #include <algorithm>
 
@@ -46,6 +48,20 @@ namespace {
 constexpr uint32_t FNV_PRIME = 0x01000193;         // 16777619
 constexpr uint32_t FNV_OFFSET_BASIS = 0x811C9DC5;  // 2166136261
 
+// On constrained targets, loading the CSS rules map before chapter parsing can
+// consume a large share of available heap and increase parse truncation risk.
+// Allow compile-time override for tuning.
+#ifndef SCT_EMBEDDED_STYLE_MIN_FREE_HEAP_BYTES
+#define SCT_EMBEDDED_STYLE_MIN_FREE_HEAP_BYTES (96 * 1024)
+#endif
+
+#ifndef SCT_EMBEDDED_STYLE_MIN_CONTIG_HEAP_BYTES
+#define SCT_EMBEDDED_STYLE_MIN_CONTIG_HEAP_BYTES (56 * 1024)
+#endif
+
+constexpr uint32_t EMBEDDED_STYLE_MIN_FREE_HEAP_BYTES = SCT_EMBEDDED_STYLE_MIN_FREE_HEAP_BYTES;
+constexpr uint32_t EMBEDDED_STYLE_MIN_CONTIG_HEAP_BYTES = SCT_EMBEDDED_STYLE_MIN_CONTIG_HEAP_BYTES;
+
 uint32_t fnv1a(const uint8_t* data, size_t length) {
   uint32_t hash = FNV_OFFSET_BASIS;
   for (size_t i = 0; i < length; ++i) {
@@ -357,6 +373,21 @@ bool Section::createSectionFile(const int fontId, const float lineCompression, c
                                 const uint16_t viewportHeight, const bool hyphenationEnabled, const bool embeddedStyle,
                                 const bool bionicReadingEnabled, const uint8_t imageRendering,
                                 const std::function<void(int)>& progressFn) {
+  if (embeddedStyle) {
+    const uint32_t freeHeap = esp_get_free_heap_size();
+    const uint32_t contigHeap = heap_caps_get_largest_free_block(MALLOC_CAP_8BIT | MALLOC_CAP_DEFAULT);
+    if (freeHeap < EMBEDDED_STYLE_MIN_FREE_HEAP_BYTES || contigHeap < EMBEDDED_STYLE_MIN_CONTIG_HEAP_BYTES) {
+      LOG_INF("SCT",
+              "Low heap for embedded CSS (free=%lu contig=%lu, need free>=%lu contig>=%lu); "
+              "building no-CSS section cache",
+              freeHeap, contigHeap, static_cast<uint32_t>(EMBEDDED_STYLE_MIN_FREE_HEAP_BYTES),
+              static_cast<uint32_t>(EMBEDDED_STYLE_MIN_CONTIG_HEAP_BYTES));
+      return createSectionFile(fontId, lineCompression, extraParagraphSpacing, paragraphAlignment, viewportWidth,
+                               viewportHeight, hyphenationEnabled, false, bionicReadingEnabled, imageRendering,
+                               progressFn);
+    }
+  }
+
   uint32_t propertyHash =
       calculatePropertyHash(fontId, lineCompression, extraParagraphSpacing, paragraphAlignment, viewportWidth,
                             viewportHeight, hyphenationEnabled, embeddedStyle, bionicReadingEnabled, imageRendering);
@@ -401,6 +432,7 @@ bool Section::createSectionFile(const int fontId, const float lineCompression, c
       if (!cssParser->loadFromCache()) {
         LOG_ERR("SCT", "Failed to load CSS from cache");
       }
+      cssParser->resetResolveStats();
     }
   }
 
@@ -440,6 +472,9 @@ bool Section::createSectionFile(const int fontId, const float lineCompression, c
   const bool streamOk = epub->readItemContentsToStream(localPath, visitor, 1024);
   const bool finalizeOk = visitor.finalize();
   const bool parserStreamOk = visitor.streamSucceeded();
+  if (cssParser) {
+    cssParser->logResolveStats(localPath.c_str());
+  }
   const bool parseComplete = streamOk && finalizeOk && parserStreamOk;
   bool success = parseComplete;
   const bool hasParsedPages = pageCount > 0;

lib/Epub/Epub/css/CssParser.cpp

@@ -45,10 +45,32 @@ constexpr size_t MAX_RULES = 1500;
 // If below this threshold, we skip CSS to avoid display artifacts.
 constexpr size_t MIN_FREE_HEAP_FOR_CSS = 48 * 1024;
 
+// In-memory CSS rule cache sizing for disk-backed lookup mode.
+// Keeps memory bounded on large books while retaining hot selectors.
+#ifndef CSS_HOT_RULE_CACHE_SIZE
+#define CSS_HOT_RULE_CACHE_SIZE 128
+#endif
+
+#ifndef CSS_NEGATIVE_CACHE_SIZE
+#define CSS_NEGATIVE_CACHE_SIZE 256
+#endif
+
+constexpr size_t HOT_RULE_CACHE_SIZE = CSS_HOT_RULE_CACHE_SIZE;
+constexpr size_t NEGATIVE_CACHE_SIZE = CSS_NEGATIVE_CACHE_SIZE;
+
 // Maximum length for a single selector string
 // Prevents parsing of extremely long or malformed selectors
 constexpr size_t MAX_SELECTOR_LENGTH = 256;
 
+constexpr size_t CSS_LENGTH_FIELD_COUNT = 11;
+constexpr size_t CSS_LENGTH_BYTES = sizeof(float) + sizeof(uint8_t);
+constexpr size_t CSS_FIXED_STYLE_BYTES =
+    4 * sizeof(uint8_t) + (CSS_LENGTH_FIELD_COUNT * CSS_LENGTH_BYTES) + sizeof(uint8_t) + sizeof(uint16_t);
+
+// Cache file name (version is CssParser::CSS_CACHE_VERSION)
+constexpr char rulesCache[] = "/css_rules.cache";
+constexpr char compileTempRulesCache[] = "/css_rules.compile.tmp";
+
 // Check if character is CSS whitespace
 bool isCssWhitespace(const char c) { return c == ' ' || c == '\t' || c == '\n' || c == '\r' || c == '\f'; }
 
@@ -445,12 +467,40 @@ void CssParser::processRuleBlockWithStyle(const std::string& selectorGroup, cons
     }
 
     // Skip if this would exceed the rule limit
-    if (rulesBySelector_.size() >= MAX_RULES) {
+    const size_t ruleCount = compileModeActive_ ? compileSelectorOffsets_.size() : rulesBySelector_.size();
+    if (ruleCount >= MAX_RULES) {
       LOG_DBG("CSS", "Reached max rules limit, stopping selector processing");
       return;
     }
 
-    // Store or merge with existing
+    if (compileModeActive_) {
+      if (!compileTempFile_) {
+        compileModeFailed_ = true;
+        continue;
+      }
+
+      CssStyle merged = style;
+      auto existingOffsetIt = compileSelectorOffsets_.find(key);
+      if (existingOffsetIt != compileSelectorOffsets_.end()) {
+        CssStyle existing;
+        FsFile tempRead;
+        if (Storage.openFileForRead("CSS", compileTempPath_, tempRead) && tempRead.seek(existingOffsetIt->second) &&
+            readCssStylePayload(tempRead, existing)) {
+          existing.applyOver(merged);
+          merged = existing;
+        }
+        if (tempRead) {
+          tempRead.close();
+        }
+      }
+
+      const uint32_t styleOffset = compileTempFile_.position();
+      writeCssStylePayload(compileTempFile_, merged);
+      compileSelectorOffsets_[key] = styleOffset;
+      continue;
+    }
+
+    // Store or merge with existing (non-compile mode)
     auto it = rulesBySelector_.find(key);
     if (it != rulesBySelector_.end()) {
       it->second.applyOver(style);
@@ -608,25 +658,432 @@ bool CssParser::loadFromStream(FsFile& source) {
   return true;
 }
 
+bool CssParser::beginCacheCompile() {
+  clear();
+  compileTempPath_ = cachePath + compileTempRulesCache;
+  Storage.remove(compileTempPath_.c_str());
+  if (!Storage.openFileForWrite("CSS", compileTempPath_, compileTempFile_)) {
+    return false;
+  }
+  compileSelectorOffsets_.clear();
+  compileModeActive_ = true;
+  compileModeFailed_ = false;
+  return true;
+}
+
+bool CssParser::appendCompiledFromStream(FsFile& source) {
+  if (!compileModeActive_) {
+    return false;
+  }
+  if (!loadFromStream(source)) {
+    compileModeFailed_ = true;
+    return false;
+  }
+  return !compileModeFailed_;
+}
+
+bool CssParser::endCacheCompile() {
+  if (!compileModeActive_) {
+    return false;
+  }
+
+  compileModeActive_ = false;
+  compileTempFile_.close();
+
+  if (compileModeFailed_) {
+    Storage.remove(compileTempPath_.c_str());
+    compileSelectorOffsets_.clear();
+    return false;
+  }
+
+  FsFile outFile;
+  if (!Storage.openFileForWrite("CSS", cachePath + rulesCache, outFile)) {
+    Storage.remove(compileTempPath_.c_str());
+    compileSelectorOffsets_.clear();
+    return false;
+  }
+
+  outFile.write(CssParser::CSS_CACHE_VERSION);
+  const auto ruleCount = static_cast<uint16_t>(compileSelectorOffsets_.size());
+  outFile.write(reinterpret_cast<const uint8_t*>(&ruleCount), sizeof(ruleCount));
+
+  FsFile tempFile;
+  if (!Storage.openFileForRead("CSS", compileTempPath_, tempFile)) {
+    outFile.close();
+    Storage.remove((cachePath + rulesCache).c_str());
+    Storage.remove(compileTempPath_.c_str());
+    compileSelectorOffsets_.clear();
+    return false;
+  }
+
+  std::array<uint8_t, CSS_FIXED_STYLE_BYTES> styleBytes{};
+  for (const auto& it : compileSelectorOffsets_) {
+    const auto selectorLen = static_cast<uint16_t>(it.first.size());
+    outFile.write(reinterpret_cast<const uint8_t*>(&selectorLen), sizeof(selectorLen));
+    outFile.write(reinterpret_cast<const uint8_t*>(it.first.data()), selectorLen);
+
+    if (!tempFile.seek(it.second)) {
+      tempFile.close();
+      outFile.close();
+      Storage.remove((cachePath + rulesCache).c_str());
+      Storage.remove(compileTempPath_.c_str());
+      compileSelectorOffsets_.clear();
+      return false;
+    }
+    if (tempFile.read(styleBytes.data(), styleBytes.size()) != static_cast<int>(styleBytes.size())) {
+      tempFile.close();
+      outFile.close();
+      Storage.remove((cachePath + rulesCache).c_str());
+      Storage.remove(compileTempPath_.c_str());
+      compileSelectorOffsets_.clear();
+      return false;
+    }
+    outFile.write(styleBytes.data(), styleBytes.size());
+  }
+
+  tempFile.close();
+  outFile.close();
+  Storage.remove(compileTempPath_.c_str());
+
+  compileSelectorOffsets_.clear();
+
+  rulesBySelector_.clear();
+  hotRuleCache_.clear();
+  hotRuleLru_.clear();
+  negativeRuleCache_.clear();
+  cacheRuleOffsets_.clear();
+  cacheIndexLoaded_ = false;
+  cachedRuleCount_ = 0;
+
+  return ensureCacheIndexLoaded();
+}
+
+bool CssParser::empty() const { return ruleCount() == 0; }
+
+size_t CssParser::ruleCount() const {
+  if (!rulesBySelector_.empty()) {
+    return rulesBySelector_.size();
+  }
+  if (cacheIndexLoaded_) {
+    return cachedRuleCount_;
+  }
+  return 0;
+}
+
+void CssParser::clear() {
+  if (compileTempFile_) {
+    compileTempFile_.close();
+  }
+  rulesBySelector_.clear();
+  cacheRuleOffsets_.clear();
+  hotRuleCache_.clear();
+  hotRuleLru_.clear();
+  negativeRuleCache_.clear();
+  cacheIndexLoaded_ = false;
+  cachedRuleCount_ = 0;
+  resolveStats_ = {};
+  compileModeActive_ = false;
+  compileModeFailed_ = false;
+  compileSelectorOffsets_.clear();
+}
+
+void CssParser::resetResolveStats() const { resolveStats_ = {}; }
+
+CssParser::ResolveStats CssParser::getResolveStats() const { return resolveStats_; }
+
+void CssParser::logResolveStats(const char* context) const {
+  const auto s = getResolveStats();
+  LOG_DBG("CSS",
+          "resolve stats[%s]: calls=%lu lowHeapSkips=%lu mapHits=%lu hotHits=%lu diskHits=%lu misses=%lu "
+          "negativeHits=%lu hotSize=%u indexSize=%u",
+          context ? context : "n/a", s.resolveCalls, s.lowHeapSkips, s.mapHits, s.hotHits, s.diskHits, s.misses,
+          s.negativeHits, static_cast<unsigned>(hotRuleCache_.size()), static_cast<unsigned>(cachedRuleCount_));
+}
+
+bool CssParser::readCssStylePayload(FsFile& file, CssStyle& style) {
+  uint8_t enumVal;
+  if (file.read(&enumVal, 1) != 1) {
+    return false;
+  }
+  style.textAlign = static_cast<CssTextAlign>(enumVal);
+
+  if (file.read(&enumVal, 1) != 1) {
+    return false;
+  }
+  style.fontStyle = static_cast<CssFontStyle>(enumVal);
+
+  if (file.read(&enumVal, 1) != 1) {
+    return false;
+  }
+  style.fontWeight = static_cast<CssFontWeight>(enumVal);
+
+  if (file.read(&enumVal, 1) != 1) {
+    return false;
+  }
+  style.textDecoration = static_cast<CssTextDecoration>(enumVal);
+
+  auto readLength = [&file](CssLength& len) -> bool {
+    if (file.read(&len.value, sizeof(len.value)) != sizeof(len.value)) {
+      return false;
+    }
+    uint8_t unitVal;
+    if (file.read(&unitVal, 1) != 1) {
+      return false;
+    }
+    len.unit = static_cast<CssUnit>(unitVal);
+    return true;
+  };
+
+  if (!readLength(style.textIndent) || !readLength(style.marginTop) || !readLength(style.marginBottom) ||
+      !readLength(style.marginLeft) || !readLength(style.marginRight) || !readLength(style.paddingTop) ||
+      !readLength(style.paddingBottom) || !readLength(style.paddingLeft) || !readLength(style.paddingRight) ||
+      !readLength(style.imageHeight) || !readLength(style.imageWidth)) {
+    return false;
+  }
+
+  uint8_t displayVal;
+  if (file.read(&displayVal, 1) != 1) {
+    return false;
+  }
+  style.display = static_cast<CssDisplay>(displayVal);
+
+  uint16_t definedBits = 0;
+  if (file.read(&definedBits, sizeof(definedBits)) != sizeof(definedBits)) {
+    return false;
+  }
+  style.defined.textAlign = (definedBits & 1 << 0) != 0;
+  style.defined.fontStyle = (definedBits & 1 << 1) != 0;
+  style.defined.fontWeight = (definedBits & 1 << 2) != 0;
+  style.defined.textDecoration = (definedBits & 1 << 3) != 0;
+  style.defined.textIndent = (definedBits & 1 << 4) != 0;
+  style.defined.marginTop = (definedBits & 1 << 5) != 0;
+  style.defined.marginBottom = (definedBits & 1 << 6) != 0;
+  style.defined.marginLeft = (definedBits & 1 << 7) != 0;
+  style.defined.marginRight = (definedBits & 1 << 8) != 0;
+  style.defined.paddingTop = (definedBits & 1 << 9) != 0;
+  style.defined.paddingBottom = (definedBits & 1 << 10) != 0;
+  style.defined.paddingLeft = (definedBits & 1 << 11) != 0;
+  style.defined.paddingRight = (definedBits & 1 << 12) != 0;
+  style.defined.imageHeight = (definedBits & 1 << 13) != 0;
+  style.defined.imageWidth = (definedBits & 1 << 14) != 0;
+  style.defined.display = (definedBits & 1 << 15) != 0;
+  return true;
+}
+
+void CssParser::writeCssStylePayload(FsFile& file, const CssStyle& style) {
+  file.write(static_cast<uint8_t>(style.textAlign));
+  file.write(static_cast<uint8_t>(style.fontStyle));
+  file.write(static_cast<uint8_t>(style.fontWeight));
+  file.write(static_cast<uint8_t>(style.textDecoration));
+
+  auto writeLength = [&file](const CssLength& len) {
+    file.write(reinterpret_cast<const uint8_t*>(&len.value), sizeof(len.value));
+    file.write(static_cast<uint8_t>(len.unit));
+  };
+
+  writeLength(style.textIndent);
+  writeLength(style.marginTop);
+  writeLength(style.marginBottom);
+  writeLength(style.marginLeft);
+  writeLength(style.marginRight);
+  writeLength(style.paddingTop);
+  writeLength(style.paddingBottom);
+  writeLength(style.paddingLeft);
+  writeLength(style.paddingRight);
+  writeLength(style.imageHeight);
+  writeLength(style.imageWidth);
+  file.write(static_cast<uint8_t>(style.display));
+
+  uint16_t definedBits = 0;
+  if (style.defined.textAlign) definedBits |= 1 << 0;
+  if (style.defined.fontStyle) definedBits |= 1 << 1;
+  if (style.defined.fontWeight) definedBits |= 1 << 2;
+  if (style.defined.textDecoration) definedBits |= 1 << 3;
+  if (style.defined.textIndent) definedBits |= 1 << 4;
+  if (style.defined.marginTop) definedBits |= 1 << 5;
+  if (style.defined.marginBottom) definedBits |= 1 << 6;
+  if (style.defined.marginLeft) definedBits |= 1 << 7;
+  if (style.defined.marginRight) definedBits |= 1 << 8;
+  if (style.defined.paddingTop) definedBits |= 1 << 9;
+  if (style.defined.paddingBottom) definedBits |= 1 << 10;
+  if (style.defined.paddingLeft) definedBits |= 1 << 11;
+  if (style.defined.paddingRight) definedBits |= 1 << 12;
+  if (style.defined.imageHeight) definedBits |= 1 << 13;
+  if (style.defined.imageWidth) definedBits |= 1 << 14;
+  if (style.defined.display) definedBits |= 1 << 15;
+  file.write(reinterpret_cast<const uint8_t*>(&definedBits), sizeof(definedBits));
+}
+
+void CssParser::touchHotRule(const std::string& selector) const {
+  auto it = hotRuleCache_.find(selector);
+  if (it == hotRuleCache_.end()) {
+    return;
+  }
+  hotRuleLru_.erase(it->second.second);
+  hotRuleLru_.push_front(selector);
+  it->second.second = hotRuleLru_.begin();
+}
+
+void CssParser::cacheHotRule(const std::string& selector, const CssStyle& style) const {
+  auto it = hotRuleCache_.find(selector);
+  if (it != hotRuleCache_.end()) {
+    it->second.first = style;
+    touchHotRule(selector);
+    return;
+  }
+
+  hotRuleLru_.push_front(selector);
+  hotRuleCache_.emplace(selector, std::make_pair(style, hotRuleLru_.begin()));
+  if (hotRuleCache_.size() > HOT_RULE_CACHE_SIZE) {
+    const std::string& evictKey = hotRuleLru_.back();
+    hotRuleCache_.erase(evictKey);
+    hotRuleLru_.pop_back();
+  }
+}
+
+bool CssParser::readRuleFromDiskAtOffset(const uint32_t styleOffset, CssStyle& outStyle) const {
+  FsFile file;
+  if (!Storage.openFileForRead("CSS", cachePath + rulesCache, file)) {
+    return false;
+  }
+  if (!file.seek(styleOffset)) {
+    file.close();
+    return false;
+  }
+  const bool ok = readCssStylePayload(file, outStyle);
+  file.close();
+  return ok;
+}
+
+bool CssParser::lookupRule(const std::string& selector, CssStyle& outStyle) const {
+  auto mapIt = rulesBySelector_.find(selector);
+  if (mapIt != rulesBySelector_.end()) {
+    outStyle = mapIt->second;
+    resolveStats_.mapHits++;
+    return true;
+  }
+
+  auto hotIt = hotRuleCache_.find(selector);
+  if (hotIt != hotRuleCache_.end()) {
+    outStyle = hotIt->second.first;
+    touchHotRule(selector);
+    resolveStats_.hotHits++;
+    return true;
+  }
+
+  if (negativeRuleCache_.find(selector) != negativeRuleCache_.end()) {
+    resolveStats_.negativeHits++;
+    return false;
+  }
+
+  if (!ensureCacheIndexLoaded()) {
+    return false;
+  }
+
+  const auto offsetIt = cacheRuleOffsets_.find(selector);
+  if (offsetIt == cacheRuleOffsets_.end()) {
+    if (negativeRuleCache_.size() >= NEGATIVE_CACHE_SIZE) {
+      negativeRuleCache_.clear();
+    }
+    negativeRuleCache_.insert(selector);
+    return false;
+  }
+
+  if (!readRuleFromDiskAtOffset(offsetIt->second, outStyle)) {
+    return false;
+  }
+
+  cacheHotRule(selector, outStyle);
+  resolveStats_.diskHits++;
+  return true;
+}
+
+bool CssParser::ensureCacheIndexLoaded() const {
+  if (cacheIndexLoaded_) {
+    return true;
+  }
+
+  FsFile file;
+  if (!Storage.openFileForRead("CSS", cachePath + rulesCache, file)) {
+    return false;
+  }
+
+  uint8_t version = 0;
+  if (file.read(&version, 1) != 1 || version != CssParser::CSS_CACHE_VERSION) {
+    file.close();
+    Storage.remove((cachePath + rulesCache).c_str());
+    return false;
+  }
+
+  uint16_t ruleCount = 0;
+  if (file.read(&ruleCount, sizeof(ruleCount)) != sizeof(ruleCount) || ruleCount > MAX_RULES) {
+    file.close();
+    return false;
+  }
+
+  cacheRuleOffsets_.clear();
+  cacheRuleOffsets_.reserve(ruleCount);
+  hotRuleCache_.clear();
+  hotRuleLru_.clear();
+  negativeRuleCache_.clear();
+
+  for (uint16_t i = 0; i < ruleCount; ++i) {
+    uint16_t selectorLen = 0;
+    if (file.read(&selectorLen, sizeof(selectorLen)) != sizeof(selectorLen) || selectorLen == 0 ||
+        selectorLen > MAX_SELECTOR_LENGTH) {
+      file.close();
+      cacheRuleOffsets_.clear();
+      return false;
+    }
+
+    std::string selector;
+    selector.resize(selectorLen);
+    if (file.read(&selector[0], selectorLen) != selectorLen) {
+      file.close();
+      cacheRuleOffsets_.clear();
+      return false;
+    }
+
+    const uint32_t styleOffset = file.position();
+    cacheRuleOffsets_[std::move(selector)] = styleOffset;
+    if (!file.seek(styleOffset + CSS_FIXED_STYLE_BYTES)) {
+      file.close();
+      cacheRuleOffsets_.clear();
+      return false;
+    }
+  }
+
+  cachedRuleCount_ = cacheRuleOffsets_.size();
+  cacheIndexLoaded_ = true;
+  file.close();
+  LOG_DBG("CSS", "Loaded CSS index: %u selectors (hot cache size=%u)", static_cast<unsigned>(cachedRuleCount_),
+          static_cast<unsigned>(HOT_RULE_CACHE_SIZE));
+  return true;
+}
+
 // Style resolution
 
 CssStyle CssParser::resolveStyle(const std::string& tagName, const std::string& classAttr) const {
   static bool lowHeapWarningLogged = false;
+  resolveStats_.resolveCalls++;
   if (ESP.getFreeHeap() < MIN_FREE_HEAP_FOR_CSS) {
     if (!lowHeapWarningLogged) {
       lowHeapWarningLogged = true;
       LOG_DBG("CSS", "Warning: low heap (%u bytes) below MIN_FREE_HEAP_FOR_CSS (%u), returning empty style",
               ESP.getFreeHeap(), static_cast<unsigned>(MIN_FREE_HEAP_FOR_CSS));
     }
+    resolveStats_.lowHeapSkips++;
     return CssStyle{};
   }
   CssStyle result;
   const std::string tag = normalized(tagName);
 
   // 1. Apply element-level style (lowest priority)
-  const auto tagIt = rulesBySelector_.find(tag);
+  {
-  if (tagIt != rulesBySelector_.end()) {
+    CssStyle tagStyle;
-    result.applyOver(tagIt->second);
+    if (lookupRule(tag, tagStyle)) {
+      result.applyOver(tagStyle);
+    }
   }
 
   // TODO: Support combinations of classes (e.g. style on .class1.class2)
@@ -637,9 +1094,9 @@ CssStyle CssParser::resolveStyle(const std::string& tagName, const std::string&
     for (const auto& cls : classes) {
       std::string classKey = "." + normalized(cls);
 
-      auto classIt = rulesBySelector_.find(classKey);
+      CssStyle classStyle;
-      if (classIt != rulesBySelector_.end()) {
+      if (lookupRule(classKey, classStyle)) {
-        result.applyOver(classIt->second);
+        result.applyOver(classStyle);
       }
     }
 
@@ -648,13 +1105,17 @@ CssStyle CssParser::resolveStyle(const std::string& tagName, const std::string&
     for (const auto& cls : classes) {
       std::string combinedKey = tag + "." + normalized(cls);
 
-      auto combinedIt = rulesBySelector_.find(combinedKey);
+      CssStyle combinedStyle;
-      if (combinedIt != rulesBySelector_.end()) {
+      if (lookupRule(combinedKey, combinedStyle)) {
-        result.applyOver(combinedIt->second);
+        result.applyOver(combinedStyle);
       }
     }
   }
 
+  if (!result.defined.anySet()) {
+    resolveStats_.misses++;
+  }
+
   return result;
 }
 
@@ -664,9 +1125,6 @@ CssStyle CssParser::parseInlineStyle(const std::string& styleValue) { return par
 
 // Cache serialization
 
-// Cache file name (version is CssParser::CSS_CACHE_VERSION)
-constexpr char rulesCache[] = "/css_rules.cache";
-
 bool CssParser::hasCache() const { return Storage.exists((cachePath + rulesCache).c_str()); }
 
 void CssParser::deleteCache() const {
@@ -754,175 +1212,19 @@ bool CssParser::loadFromCache() {
     return false;
   }
 
-  FsFile file;
+  // Drop parse-time in-memory rules, then initialize on-disk selector index.
-  if (!Storage.openFileForRead("CSS", cachePath + rulesCache, file)) {
-    return false;
-  }
-
-  // Clear existing rules
-  clear();
-
-  // Read and verify version
-  uint8_t version = 0;
-  if (file.read(&version, 1) != 1 || version != CssParser::CSS_CACHE_VERSION) {
-    LOG_DBG("CSS", "Cache version mismatch (got %u, expected %u), removing stale cache for rebuild", version,
-            CssParser::CSS_CACHE_VERSION);
-    file.close();
-    Storage.remove((cachePath + rulesCache).c_str());
-    return false;
-  }
-
-  // Read rule count
-  uint16_t ruleCount = 0;
-  if (file.read(&ruleCount, sizeof(ruleCount)) != sizeof(ruleCount)) {
-    file.close();
-    return false;
-  }
-
-  if (ruleCount > MAX_RULES) {
-    LOG_DBG("CSS", "Invalid cache rule count (%u > %zu)", ruleCount, MAX_RULES);
   rulesBySelector_.clear();
-    file.close();
+  hotRuleCache_.clear();
+  hotRuleLru_.clear();
+  negativeRuleCache_.clear();
+  cacheRuleOffsets_.clear();
+  cacheIndexLoaded_ = false;
+  cachedRuleCount_ = 0;
+
+  if (!ensureCacheIndexLoaded()) {
     return false;
   }
 
-  auto hasRemainingBytes = [&file](const size_t neededBytes) -> bool {
+  LOG_DBG("CSS", "Loaded %u rules from cache index", static_cast<unsigned>(cachedRuleCount_));
-    return static_cast<size_t>(file.available()) >= neededBytes;
-  };
-
-  constexpr size_t CSS_LENGTH_FIELD_COUNT = 11;
-  constexpr size_t CSS_LENGTH_BYTES = sizeof(float) + sizeof(uint8_t);
-  constexpr size_t CSS_FIXED_STYLE_BYTES =
-      4 * sizeof(uint8_t) + (CSS_LENGTH_FIELD_COUNT * CSS_LENGTH_BYTES) + sizeof(uint8_t) + sizeof(uint16_t);
-
-  // Read each rule
-  for (uint16_t i = 0; i < ruleCount; ++i) {
-    // Read selector string
-    uint16_t selectorLen = 0;
-    if (!hasRemainingBytes(sizeof(selectorLen))) {
-      rulesBySelector_.clear();
-      file.close();
-      return false;
-    }
-    if (file.read(&selectorLen, sizeof(selectorLen)) != sizeof(selectorLen)) {
-      rulesBySelector_.clear();
-      file.close();
-      return false;
-    }
-
-    if (selectorLen == 0 || selectorLen > MAX_SELECTOR_LENGTH || !hasRemainingBytes(selectorLen)) {
-      LOG_DBG("CSS", "Invalid selector length in cache: %u", selectorLen);
-      rulesBySelector_.clear();
-      file.close();
-      return false;
-    }
-
-    std::string selector;
-    selector.resize(selectorLen);
-    if (file.read(&selector[0], selectorLen) != selectorLen) {
-      rulesBySelector_.clear();
-      file.close();
-      return false;
-    }
-
-    if (!hasRemainingBytes(CSS_FIXED_STYLE_BYTES)) {
-      LOG_DBG("CSS", "Truncated CSS cache while reading style payload");
-      rulesBySelector_.clear();
-      file.close();
-      return false;
-    }
-
-    // Read CssStyle fields
-    CssStyle style;
-    uint8_t enumVal;
-
-    if (file.read(&enumVal, 1) != 1) {
-      rulesBySelector_.clear();
-      file.close();
-      return false;
-    }
-    style.textAlign = static_cast<CssTextAlign>(enumVal);
-
-    if (file.read(&enumVal, 1) != 1) {
-      rulesBySelector_.clear();
-      file.close();
-      return false;
-    }
-    style.fontStyle = static_cast<CssFontStyle>(enumVal);
-
-    if (file.read(&enumVal, 1) != 1) {
-      rulesBySelector_.clear();
-      file.close();
-      return false;
-    }
-    style.fontWeight = static_cast<CssFontWeight>(enumVal);
-
-    if (file.read(&enumVal, 1) != 1) {
-      rulesBySelector_.clear();
-      file.close();
-      return false;
-    }
-    style.textDecoration = static_cast<CssTextDecoration>(enumVal);
-
-    // Read CssLength fields
-    auto readLength = [&file](CssLength& len) -> bool {
-      if (file.read(&len.value, sizeof(len.value)) != sizeof(len.value)) {
-        return false;
-      }
-      uint8_t unitVal;
-      if (file.read(&unitVal, 1) != 1) {
-        return false;
-      }
-      len.unit = static_cast<CssUnit>(unitVal);
-      return true;
-    };
-
-    if (!readLength(style.textIndent) || !readLength(style.marginTop) || !readLength(style.marginBottom) ||
-        !readLength(style.marginLeft) || !readLength(style.marginRight) || !readLength(style.paddingTop) ||
-        !readLength(style.paddingBottom) || !readLength(style.paddingLeft) || !readLength(style.paddingRight) ||
-        !readLength(style.imageHeight) || !readLength(style.imageWidth)) {
-      rulesBySelector_.clear();
-      file.close();
-      return false;
-    }
-
-    // Read display value
-    uint8_t displayVal;
-    if (file.read(&displayVal, 1) != 1) {
-      rulesBySelector_.clear();
-      file.close();
-      return false;
-    }
-    style.display = static_cast<CssDisplay>(displayVal);
-
-    // Read defined flags
-    uint16_t definedBits = 0;
-    if (file.read(&definedBits, sizeof(definedBits)) != sizeof(definedBits)) {
-      rulesBySelector_.clear();
-      file.close();
-      return false;
-    }
-    style.defined.textAlign = (definedBits & 1 << 0) != 0;
-    style.defined.fontStyle = (definedBits & 1 << 1) != 0;
-    style.defined.fontWeight = (definedBits & 1 << 2) != 0;
-    style.defined.textDecoration = (definedBits & 1 << 3) != 0;
-    style.defined.textIndent = (definedBits & 1 << 4) != 0;
-    style.defined.marginTop = (definedBits & 1 << 5) != 0;
-    style.defined.marginBottom = (definedBits & 1 << 6) != 0;
-    style.defined.marginLeft = (definedBits & 1 << 7) != 0;
-    style.defined.marginRight = (definedBits & 1 << 8) != 0;
-    style.defined.paddingTop = (definedBits & 1 << 9) != 0;
-    style.defined.paddingBottom = (definedBits & 1 << 10) != 0;
-    style.defined.paddingLeft = (definedBits & 1 << 11) != 0;
-    style.defined.paddingRight = (definedBits & 1 << 12) != 0;
-    style.defined.imageHeight = (definedBits & 1 << 13) != 0;
-    style.defined.imageWidth = (definedBits & 1 << 14) != 0;
-    style.defined.display = (definedBits & 1 << 15) != 0;
-
-    rulesBySelector_[selector] = style;
-  }
-
-  LOG_DBG("CSS", "Loaded %u rules from cache", ruleCount);
-  file.close();
   return true;
 }

lib/Epub/Epub/css/CssParser.h

@@ -2,8 +2,10 @@
 
 #include <HalStorage.h>
 
+#include <list>
 #include <string>
 #include <unordered_map>
+#include <unordered_set>
 #include <utility>
 #include <vector>
 
@@ -30,6 +32,16 @@
  */
 class CssParser {
  public:
+  struct ResolveStats {
+    uint32_t resolveCalls = 0;
+    uint32_t lowHeapSkips = 0;
+    uint32_t mapHits = 0;
+    uint32_t hotHits = 0;
+    uint32_t diskHits = 0;
+    uint32_t misses = 0;
+    uint32_t negativeHits = 0;
+  };
+
   // Bump when CSS cache format or rules change; section caches are invalidated when this changes
   static constexpr uint8_t CSS_CACHE_VERSION = 4;
 
@@ -68,17 +80,17 @@ class CssParser {
   /**
    * Check if any rules have been loaded
    */
-  [[nodiscard]] bool empty() const { return rulesBySelector_.empty(); }
+  [[nodiscard]] bool empty() const;
 
   /**
    * Get count of loaded rule sets
    */
-  [[nodiscard]] size_t ruleCount() const { return rulesBySelector_.size(); }
+  [[nodiscard]] size_t ruleCount() const;
 
   /**
    * Clear all loaded rules
    */
-  void clear() { rulesBySelector_.clear(); }
+  void clear();
 
   /**
    * Check if CSS rules cache file exists
@@ -103,12 +115,43 @@ class CssParser {
    */
   bool loadFromCache();
 
+  // Low-memory CSS compilation pipeline:
+  // - beginCacheCompile(): starts streaming compile mode
+  // - appendCompiledFromStream(): parses one stylesheet stream into compile staging
+  // - endCacheCompile(): finalizes cache file from staged records
+  bool beginCacheCompile();
+  bool appendCompiledFromStream(FsFile& source);
+  bool endCacheCompile();
+
+  // CSS lookup telemetry helpers for tuning memory/caching behavior on-device.
+  void resetResolveStats() const;
+  [[nodiscard]] ResolveStats getResolveStats() const;
+  void logResolveStats(const char* context) const;
+
  private:
   // Storage: maps normalized selector -> style properties
   std::unordered_map<std::string, CssStyle> rulesBySelector_;
 
   std::string cachePath;
 
+  // Disk-backed CSS dictionary index: selector -> byte offset for serialized CssStyle payload.
+  // Built from cache file once, then styles are loaded on demand into hotRuleCache_.
+  mutable bool cacheIndexLoaded_ = false;
+  mutable size_t cachedRuleCount_ = 0;
+  mutable std::unordered_map<std::string, uint32_t> cacheRuleOffsets_;
+
+  // Bounded hot cache of most recently used rules.
+  mutable std::list<std::string> hotRuleLru_;
+  mutable std::unordered_map<std::string, std::pair<CssStyle, std::list<std::string>::iterator>> hotRuleCache_;
+  mutable std::unordered_set<std::string> negativeRuleCache_;
+  mutable ResolveStats resolveStats_;
+
+  bool compileModeActive_ = false;
+  bool compileModeFailed_ = false;
+  std::string compileTempPath_;
+  FsFile compileTempFile_;
+  std::unordered_map<std::string, uint32_t> compileSelectorOffsets_;
+
   // Internal parsing helpers
   void processRuleBlockWithStyle(const std::string& selectorGroup, const CssStyle& style);
   static CssStyle parseDeclarations(const std::string& declBlock);
@@ -129,4 +172,13 @@ class CssParser {
   static void normalizedInto(const std::string& s, std::string& out);
   static std::vector<std::string> splitOnChar(const std::string& s, char delimiter);
   static std::vector<std::string> splitWhitespace(const std::string& s);
+
+  // On-demand rule loading helpers
+  bool ensureCacheIndexLoaded() const;
+  bool lookupRule(const std::string& selector, CssStyle& outStyle) const;
+  bool readRuleFromDiskAtOffset(uint32_t styleOffset, CssStyle& outStyle) const;
+  static bool readCssStylePayload(FsFile& file, CssStyle& style);
+  static void writeCssStylePayload(FsFile& file, const CssStyle& style);
+  void touchHotRule(const std::string& selector) const;
+  void cacheHotRule(const std::string& selector, const CssStyle& style) const;
 };

lib/GfxRenderer/GfxRenderer.cpp

@@ -69,6 +69,9 @@ void GfxRenderer::begin() {
   panelHeight = display.getDisplayHeight();
   panelWidthBytes = display.getDisplayWidthBytes();
   frameBufferSize = display.getBufferSize();
+  bwSnapshotRowStart = 0;
+  bwSnapshotRowEnd = 0;
+  bwSnapshotSizeBytes = 0;
   bwBufferChunkSize = BW_BUFFER_CHUNK_SIZE;
   bwBufferChunks.assign((frameBufferSize + bwBufferChunkSize - 1) / bwBufferChunkSize, nullptr);
 }
@@ -2143,9 +2146,75 @@ void GfxRenderer::freeBwBufferChunks() {
  * Uses chunked allocation to avoid needing 48KB of contiguous memory.
  * Returns true if buffer was stored successfully, false if allocation failed.
  */
-bool GfxRenderer::storeBwBuffer() {
+bool GfxRenderer::storeBwBuffer() { return storeBwBufferRect(0, 0, getScreenWidth(), getScreenHeight()); }
+
+bool GfxRenderer::storeBwBufferRect(const int x, const int y, const int width, const int height) {
+  if (width <= 0 || height <= 0) {
+    LOG_ERR("GFX", "!! BW buffer store rect invalid: x=%d y=%d w=%d h=%d", x, y, width, height);
+    return false;
+  }
+
+  const int screenWidth = getScreenWidth();
+  const int screenHeight = getScreenHeight();
+  if (screenWidth <= 0 || screenHeight <= 0 || panelWidthBytes == 0 || panelHeight == 0 || !frameBuffer) {
+    LOG_ERR("GFX", "!! BW buffer store unavailable (screen=%dx%d panelHeight=%u rowBytes=%u fb=%p)", screenWidth,
+            screenHeight, panelHeight, panelWidthBytes, frameBuffer);
+    return false;
+  }
+
+  const int clampedX0 = std::max(0, x);
+  const int clampedY0 = std::max(0, y);
+  const int clampedX1 = std::min(screenWidth - 1, x + width - 1);
+  const int clampedY1 = std::min(screenHeight - 1, y + height - 1);
+  if (clampedX0 > clampedX1 || clampedY0 > clampedY1) {
+    LOG_ERR("GFX", "!! BW buffer store rect outside screen: x=%d y=%d w=%d h=%d", x, y, width, height);
+    return false;
+  }
+
+  int rowStart = 0;
+  int rowEnd = 0;
+  switch (getOrientation()) {
+    case LandscapeCounterClockwise:
+      rowStart = clampedY0;
+      rowEnd = clampedY1;
+      break;
+    case LandscapeClockwise:
+      rowStart = static_cast<int>(panelHeight) - 1 - clampedY1;
+      rowEnd = static_cast<int>(panelHeight) - 1 - clampedY0;
+      break;
+    case Portrait:
+      rowStart = static_cast<int>(panelHeight) - 1 - clampedX1;
+      rowEnd = static_cast<int>(panelHeight) - 1 - clampedX0;
+      break;
+    case PortraitInverted:
+      rowStart = clampedX0;
+      rowEnd = clampedX1;
+      break;
+  }
+
+  rowStart = std::max(0, rowStart);
+  rowEnd = std::min(static_cast<int>(panelHeight) - 1, rowEnd);
+  if (rowStart > rowEnd) {
+    LOG_ERR("GFX", "!! BW buffer store row-band invalid after orientation mapping: rows=%d..%d", rowStart, rowEnd);
+    return false;
+  }
+
+  const size_t rows = static_cast<size_t>(rowEnd - rowStart + 1);
+  const size_t snapshotSizeBytes = rows * panelWidthBytes;
+  const size_t snapshotBaseOffset = static_cast<size_t>(rowStart) * panelWidthBytes;
+  if (snapshotSizeBytes == 0 || snapshotBaseOffset + snapshotSizeBytes > frameBufferSize) {
+    LOG_ERR("GFX", "!! BW buffer store row-band out of bounds: base=%zu size=%zu frame=%u", snapshotBaseOffset,
+            snapshotSizeBytes, frameBufferSize);
+    return false;
+  }
+
+  freeBwBufferChunks();
+  bwSnapshotRowStart = static_cast<uint16_t>(rowStart);
+  bwSnapshotRowEnd = static_cast<uint16_t>(rowEnd);
+  bwSnapshotSizeBytes = snapshotSizeBytes;
+
   auto attemptStore = [&](size_t chunkSize) {
-    bwBufferChunks.assign((frameBufferSize + chunkSize - 1) / chunkSize, nullptr);
+    bwBufferChunks.assign((bwSnapshotSizeBytes + chunkSize - 1) / chunkSize, nullptr);
     for (size_t i = 0; i < bwBufferChunks.size(); i++) {
       if (bwBufferChunks[i]) {
         LOG_ERR("GFX", "!! BW buffer chunk %zu already stored - this is likely a bug, freeing chunk", i);
@@ -2154,7 +2223,7 @@ bool GfxRenderer::storeBwBuffer() {
       }
 
       const size_t offset = i * chunkSize;
-      const size_t allocSize = std::min(chunkSize, static_cast<size_t>(frameBufferSize - offset));
+      const size_t allocSize = std::min(chunkSize, bwSnapshotSizeBytes - offset);
       bwBufferChunks[i] = static_cast<uint8_t*>(malloc(allocSize));
 
       if (!bwBufferChunks[i]) {
@@ -2166,10 +2235,11 @@ bool GfxRenderer::storeBwBuffer() {
         return false;
       }
 
-      memcpy(bwBufferChunks[i], frameBuffer + offset, allocSize);
+      memcpy(bwBufferChunks[i], frameBuffer + snapshotBaseOffset + offset, allocSize);
     }
     bwBufferChunkSize = chunkSize;
-    LOG_DBG("GFX", "Stored BW buffer in %zu chunks (%zu bytes each)", bwBufferChunks.size(), chunkSize);
+    LOG_DBG("GFX", "Stored BW buffer rows [%u..%u] (%zu bytes) in %zu chunks (%zu bytes each)", bwSnapshotRowStart,
+            bwSnapshotRowEnd, bwSnapshotSizeBytes, bwBufferChunks.size(), chunkSize);
     return true;
   };
 
@@ -2199,6 +2269,9 @@ bool GfxRenderer::storeBwBuffer() {
   }
 
   LOG_ERR("GFX", "!! BW buffer storage failed after retrying smaller chunk sizes");
+  bwSnapshotSizeBytes = 0;
+  bwSnapshotRowStart = 0;
+  bwSnapshotRowEnd = 0;
   return false;
 }
 
@@ -2208,6 +2281,13 @@ bool GfxRenderer::storeBwBuffer() {
  * Uses chunked restoration to match chunked storage.
  */
 void GfxRenderer::restoreBwBuffer() {
+  if (bwSnapshotSizeBytes == 0) {
+    display.cleanupGrayscaleBuffers(frameBuffer);
+    freeBwBufferChunks();
+    LOG_ERR("GFX", "BW restore skipped: no stored snapshot metadata; cleaned grayscale buffers only");
+    return;
+  }
+
   // Check if all chunks are allocated
   bool missingChunks = false;
   for (const auto& bwBufferChunk : bwBufferChunks) {
@@ -2223,20 +2303,28 @@ void GfxRenderer::restoreBwBuffer() {
     // allocations that can later starve TLS handshakes.
     display.cleanupGrayscaleBuffers(frameBuffer);
     freeBwBufferChunks();
+    bwSnapshotSizeBytes = 0;
+    bwSnapshotRowStart = 0;
+    bwSnapshotRowEnd = 0;
     LOG_ERR("GFX", "BW restore skipped due to missing chunks; cleaned grayscale buffers only");
     return;
   }
 
+  const size_t snapshotBaseOffset = static_cast<size_t>(bwSnapshotRowStart) * panelWidthBytes;
   for (size_t i = 0; i < bwBufferChunks.size(); i++) {
     const size_t offset = i * bwBufferChunkSize;
-    const size_t chunkSize = std::min(bwBufferChunkSize, static_cast<size_t>(frameBufferSize - offset));
+    const size_t chunkSize = std::min(bwBufferChunkSize, bwSnapshotSizeBytes - offset);
-    memcpy(frameBuffer + offset, bwBufferChunks[i], chunkSize);
+    memcpy(frameBuffer + snapshotBaseOffset + offset, bwBufferChunks[i], chunkSize);
   }
 
   display.cleanupGrayscaleBuffers(frameBuffer);
 
   freeBwBufferChunks();
-  LOG_DBG("GFX", "Restored and freed BW buffer chunks");
+  LOG_DBG("GFX", "Restored BW buffer rows [%u..%u] (%zu bytes) and freed BW chunks", bwSnapshotRowStart,
+          bwSnapshotRowEnd, bwSnapshotSizeBytes);
+  bwSnapshotSizeBytes = 0;
+  bwSnapshotRowStart = 0;
+  bwSnapshotRowEnd = 0;
 }
 
 /**

lib/GfxRenderer/GfxRenderer.h

@@ -55,6 +55,9 @@ class GfxRenderer {
   uint16_t panelHeight = 0;      // set in begin()
   uint16_t panelWidthBytes = 0;  // set in begin()
   uint32_t frameBufferSize = 0;  // set in begin()
+  uint16_t bwSnapshotRowStart = 0;
+  uint16_t bwSnapshotRowEnd = 0;
+  size_t bwSnapshotSizeBytes = 0;
   size_t bwBufferChunkSize = BW_BUFFER_CHUNK_SIZE;
   std::vector<uint8_t*> bwBufferChunks;
   std::map<int, EpdFontFamily> fontMap;
@@ -207,6 +210,7 @@ class GfxRenderer {
   void copyGrayscaleMsbBuffers() const;
   void displayGrayBuffer() const;
   bool storeBwBuffer();                                         // Returns true if buffer was stored successfully
+  bool storeBwBufferRect(int x, int y, int width, int height);  // Store only rows intersecting logical rect
   void restoreBwBuffer();                                       // Restore and free the stored buffer
   void cleanupGrayscaleWithFrameBuffer() const;
 

src/activities/reader/EpubReaderActivity.cpp

@@ -75,11 +75,11 @@ constexpr uint32_t AA_RECOVERY_CONTIG_HEAP_BYTES = CP_AA_RECOVERY_CONTIG_HEAP_BY
 // Snapshotting BW buffer for grayscale restore can fragment heap heavily on tight
 // pages. Skip attempting snapshot altogether below this safety window.
 #ifndef CP_BW_SNAPSHOT_MIN_FREE_HEAP_BYTES
-#define CP_BW_SNAPSHOT_MIN_FREE_HEAP_BYTES (72 * 1024)
+#define CP_BW_SNAPSHOT_MIN_FREE_HEAP_BYTES (64 * 1024)
 #endif
 
 #ifndef CP_BW_SNAPSHOT_MIN_CONTIG_HEAP_BYTES
-#define CP_BW_SNAPSHOT_MIN_CONTIG_HEAP_BYTES (52 * 1024)
+#define CP_BW_SNAPSHOT_MIN_CONTIG_HEAP_BYTES (24 * 1024)
 #endif
 
 constexpr uint32_t BW_SNAPSHOT_MIN_FREE_HEAP_BYTES = CP_BW_SNAPSHOT_MIN_FREE_HEAP_BYTES;
@@ -98,6 +98,62 @@ void logReaderMemSnapshot(const char* stage) {
 inline void logReaderMemSnapshot(const char*) {}
 #endif
 
+bool computePageDynamicYBand(const Page& page, const GfxRenderer& renderer, const int fontId, const int viewportHeight,
+                             int* outTop, int* outBottom) {
+  if (viewportHeight <= 0 || !outTop || !outBottom) {
+    return false;
+  }
+
+  bool hasRange = false;
+  int minY = viewportHeight;
+  int maxY = -1;
+  const int lineHeight = std::max(1, renderer.getLineHeight(fontId));
+
+  for (const auto& el : page.elements) {
+    if (!el) continue;
+
+    int elementTop = el->yPos;
+    int elementBottom = el->yPos;
+    switch (el->getTag()) {
+      case TAG_PageLine:
+        elementBottom = el->yPos + lineHeight;
+        break;
+      case TAG_PageImage: {
+        const auto& img = static_cast<const PageImage&>(*el);
+        elementBottom = el->yPos + img.getImageBlock().getHeight();
+        break;
+      }
+      case TAG_PageTable: {
+        const auto& table = static_cast<const PageTableFragment&>(*el);
+        elementBottom = el->yPos + table.getTotalHeight();
+        break;
+      }
+      default:
+        continue;
+    }
+
+    minY = std::min(minY, elementTop);
+    maxY = std::max(maxY, elementBottom);
+    hasRange = true;
+  }
+
+  if (!hasRange) {
+    return false;
+  }
+
+  constexpr int BAND_PAD_PX = 2;
+  minY = std::max(0, minY - BAND_PAD_PX);
+  maxY = std::min(viewportHeight, maxY + BAND_PAD_PX);
+
+  if (minY >= maxY) {
+    return false;
+  }
+
+  *outTop = minY;
+  *outBottom = maxY;
+  return true;
+}
+
 // Computes the [0..100] EPUB progress percent. Returns 0 when pageCount is unknown (sync/bookmark
 // pre-render writes), in which case the next saveProgress() will overwrite progress.bin with the
 // real value before the user can leave the reader.
@@ -1708,7 +1764,20 @@ void EpubReaderActivity::renderContents(std::unique_ptr<Page> page, const int or
   logReaderMemSnapshot("bw_store_begin");
   bool bwBufferStored = false;
   if (shouldAttemptBwSnapshot) {
-    bwBufferStored = renderer.storeBwBuffer();
+    const int contentLeft = orientedMarginLeft;
+    const int contentRight = std::max(contentLeft, renderer.getScreenWidth() - orientedMarginRight);
+    const int contentBottom = std::max(contentTop, renderer.getScreenHeight() - orientedMarginBottom);
+
+    int bandTop = 0;
+    int bandBottom = std::max(0, contentBottom - contentTop);
+    if (!computePageDynamicYBand(*page, renderer, getEffectiveReaderFontId(), bandBottom, &bandTop, &bandBottom)) {
+      bandTop = 0;
+      bandBottom = std::max(0, contentBottom - contentTop);
+    }
+
+    const int snapshotTop = contentTop + bandTop;
+    const int snapshotHeight = std::max(0, bandBottom - bandTop);
+    bwBufferStored = renderer.storeBwBufferRect(contentLeft, snapshotTop, contentRight - contentLeft, snapshotHeight);
   } else {
     LOG_INF("ERS", "Skipping BW snapshot precheck (free=%lu contig=%lu, need free>=%lu contig>=%lu)", bwStoreFreeHeap,
             bwStoreContigHeap, static_cast<uint32_t>(BW_SNAPSHOT_MIN_FREE_HEAP_BYTES),